Method, device and equipment for identifying high density sign of middle cerebral artery and storage medium

Abstract

The application discloses a middle cerebral artery high-density sign identification method, device, equipment and storage medium. In the identification method, first, the middle cerebral artery region of a brain CT plain scan image is extracted to obtain a region extraction image; then, according to a blood vessel enhanced image corresponding to the region extraction image, position information of a middle cerebral artery high-density sign candidate box is obtained; then, texture features are obtained according to the region extraction image and the position information, and shape features are obtained according to the blood vessel enhanced image and the position information; finally, a middle cerebral artery high-density sign identification result is obtained according to the texture features and the shape features. The application can realize accurate and reliable identification of HMCAS in a complex brain lesion environment, and has strong adaptability to scanning equipment and individual imaging differences.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a middle cerebral artery hyperdensity sign identification method, device, equipment and storage medium. BACKGROUND

[0002] The cerebral arteries include the basilar artery, vertebral artery, internal carotid artery, anterior cerebral artery, middle cerebral artery and posterior cerebral artery. Among them, the middle cerebral artery directly continues the ipsilateral internal carotid artery and participates in the blood supply of the lateral hemisphere, insular lobe, basal ganglia and thalamus of the brain. In the cerebral arteries of biological individuals, there may be an increase in the density of the contents of the blood vessels or an increase in the hematocrit in the blood vessels, forming a middle cerebral artery hyperdensity sign (HMCAS).

[0003] CT plain scan images (also known as NCCT images) are often used to observe whether there is HMCAS. Doctors need rich experience in image examination to detect HMCAS by eye, and it is time-consuming, low in efficiency and easy to misdiagnose due to fatigue. It has become a trend in the medical field to automatically detect and identify HMCAS from NCCT images.

[0004] The current common HMCAS identification method is to distinguish based on the shape of the brain and the CT value range. The shape of the HMCAS lesion is variable, and the HMCAS candidate region is easily positioned to deviate based on the shape and position information of the brain. The CT value is easily affected by the scanning device, brain shape and human body differences, and fluctuates. The judgment method based on the CT threshold cannot adapt to the scanning device and individual imaging differences. Therefore, the current HMCAS identification method has poor adaptability to scanning devices and individual imaging differences, and it is difficult to obtain a high-accuracy identification result.

[0005] In addition, the mispositioning of the head, brain edema, small nodular lesions, brain calcification or other high-density noise in the image also easily interferes with the identification of HMCAS, affecting the accuracy and reliability of the identification. It can be seen that there is a need to provide an HMCAS identification scheme that can adapt to complex brain lesion environments and avoid failing to normally identify HMCAS when accompanied by other complex brain lesions. SUMMARY

[0006] Based on the above problems, the present application provides a middle cerebral artery hyperdensity sign identification method, device, equipment and storage medium to improve the accuracy and reliability of the identification of the middle cerebral artery hyperdensity sign.

[0007] The embodiments of the present application disclose the following technical solutions:

[0008] The first aspect of the application provides a middle cerebral artery high-density sign recognition method. The recognition method comprises:

[0009] extracting a middle cerebral artery region of the brain CT plain scan image to obtain a region extraction image;

[0010] obtaining the position information of the middle cerebral artery high-density sign candidate box according to the blood vessel enhanced image corresponding to the region extraction image;

[0011] obtaining texture features according to the region extraction image and the position information, and obtaining shape features according to the blood vessel enhanced image and the position information;

[0012] obtaining a middle cerebral artery high-density sign recognition result according to the texture features and the shape features.

[0013] In an optional implementation, the obtaining of the texture features according to the region extraction image and the position information, and the obtaining of the shape features according to the blood vessel enhanced image and the position information specifically comprises:

[0014] obtaining a candidate middle cerebral artery high-density sign image according to the region extraction image and the position information, and obtaining a candidate blood vessel enhanced image according to the blood vessel enhanced image and the position information;

[0015] extracting texture features from the candidate middle cerebral artery high-density sign image, and extracting shape features from the candidate blood vessel enhanced image.

[0016] In an optional implementation, the obtaining of the middle cerebral artery high-density sign recognition result according to the texture features and the shape features specifically comprises:

[0017] obtaining features of multiple scales of the candidate middle cerebral artery high-density sign image;

[0018] merging the features of multiple scales after fusion with the texture features and the shape features to obtain merged features;

[0019] obtaining a recognition result of the candidate middle cerebral artery high-density sign image according to the merged features, the recognition result being having a middle cerebral artery high-density sign or not having a middle cerebral artery high-density sign.

[0020] In an optional implementation, the middle cerebral artery high-density sign recognition method further comprises:

[0021] segmenting the candidate middle cerebral artery high-density sign image recognized as having a middle cerebral artery high-density sign using a blood vessel mask to obtain a middle cerebral artery high-density sign segmentation result.

[0022] In an optional implementation, the texture feature is obtained according to the region extraction image and the position information, and specifically includes:

[0023] The gray level co-occurrence matrix corresponding to the middle cerebral artery high-density sign candidate box is calculated from each layer image of the region extraction image at different preset angles.

[0024] The target gray level co-occurrence matrix is obtained by averaging the calculated multiple gray level co-occurrence matrices.

[0025] The texture feature is constructed according to the target gray level co-occurrence matrix.

[0026] In an optional implementation, the texture feature is constructed according to the target gray level co-occurrence matrix, and specifically includes:

[0027] The contrast, cross-correlation, energy and homogeneity are calculated as the texture feature according to the target gray level co-occurrence matrix.

[0028] In an optional implementation, the middle cerebral artery region of the brain CT scan image is extracted to obtain a region extraction image, and specifically includes:

[0029] The target brain image is obtained by removing the non-brain parenchyma region of the brain CT scan image through a brain mask.

[0030] The candidate middle cerebral artery region label is obtained by registering the target brain image with reference to the brain template image and the middle cerebral artery region label of the brain template image.

[0031] The region extraction image is obtained by region extraction of the target brain image through the candidate middle cerebral artery region label.

[0032] In an optional implementation, the target brain image is obtained by removing the non-brain parenchyma region of the brain CT scan image through a brain mask, and specifically includes:

[0033] The brain CT scan image is denoised to obtain a processed brain CT scan image.

[0034] The brain mask is extracted from the processed brain CT scan image.

[0035] The target brain image is obtained by removing the non-brain parenchyma region of the processed brain CT scan image through the brain mask.

[0036] In an optional implementation, the position information of the middle cerebral artery high-density sign candidate box is obtained according to the blood vessel enhanced image corresponding to the region extraction image, and specifically includes:

[0037] The region extraction image is subjected to blood vessel enhancement in a filtering manner to obtain a blood vessel enhanced image corresponding to the region extraction image;

[0038] A blood vessel segmentation threshold is determined according to the distribution of the gray value in the blood vessel enhanced image;

[0039] The blood vessel enhanced image is processed according to the blood vessel segmentation threshold to obtain a blood vessel mask image;

[0040] A minimum circumscribed rectangle corresponding to each connected region in the blood vessel mask image is determined as a middle cerebral artery high density sign candidate box;

[0041] Position information of the middle cerebral artery high density sign candidate box in the image is obtained.

[0042] In an optional implementation, the blood vessel segmentation threshold is determined according to the distribution of the gray value in the blood vessel enhanced image, specifically including:

[0043] A distribution curve of the gray value is calculated according to the gray value of the pixel points other than the pixel points with a gray value of 0 in the blood vessel enhanced image; the horizontal axis of the distribution curve represents the gray value, and the vertical axis represents the number of pixel points;

[0044] The maximum value of the distribution curve on the vertical axis is determined;

[0045] A reference number is determined according to a preset coefficient and the maximum value;

[0046] The smallest gray value corresponding to the reference number is determined from the distribution curve as the blood vessel segmentation threshold.

[0047] The second aspect of the present application provides a middle cerebral artery high density sign recognition device. The recognition device includes:

[0048] A region extraction module is configured to extract a middle artery region of a brain CT plain scan image to obtain a region extraction image;

[0049] A candidate box position acquisition module is configured to obtain position information of a middle cerebral artery high density sign candidate box according to a blood vessel enhanced image corresponding to the region extraction image;

[0050] A feature extraction module is configured to obtain a texture feature according to the region extraction image and the position information, and obtain a shape feature according to the blood vessel enhanced image and the position information;

[0051] A recognition module is configured to obtain a middle cerebral artery high density sign recognition result according to the texture feature and the shape feature.

[0052] The third aspect of the present application provides a middle cerebral artery high density sign recognition device. The recognition device includes:

[0053] a memory having stored thereon a computer program;

[0054] a processor configured to execute the computer program in the memory to implement the steps of the method for identifying high middle cerebral artery sign according to the first aspect.

[0055] The fourth aspect of the present application provides a computer readable storage medium having stored thereon a computer program. The program, when executed by a processor, implements the steps of the method for identifying high middle cerebral artery sign according to the first aspect.

[0056] Compared with the prior art, the present application has the following beneficial effects:

[0057] In the method for identifying high middle cerebral artery sign provided by the present application, the middle cerebral artery region of a brain CT plain scan image is first extracted to obtain a region extraction image; then, the position information of the HMCAS candidate box is obtained according to the blood vessel enhanced image corresponding to the region extraction image; then, the texture feature is obtained according to the region extraction image and the position information, and the shape feature is obtained according to the blood vessel enhanced image and the position information; finally, the identification result of the high middle cerebral artery sign is obtained according to the texture feature and the shape feature.

[0058] In the present application, the position information of the HMCAS candidate box is obtained on the basis of the blood vessel enhanced image corresponding to the region extraction image, so the position of the candidate box has strong position correspondence with the middle cerebral artery region and the related blood vessels. The position can assist in achieving accurate HMCAS identification. In addition, the texture feature and the shape feature are extracted based on the position of the candidate box on the region extraction image and the blood vessel enhanced image, respectively. The texture feature can reflect the degree of change of the texture of the related part in the image, and the shape feature can reflect the shape characteristics of the blood vessels in the image. The combination of the two can distinguish the HMCAS from other lesions or influencing factors in the image, so as to achieve accurate and reliable identification of HMCAS in a complex brain lesion environment. In addition, this scheme is not simply based on CT value for HMCAS identification, so it has strong adaptability to scanning equipment and individual imaging differences. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0060] Figure 1A flowchart illustrating a method for identifying high-density signs in the middle cerebral artery, as provided in this application embodiment;

[0061] Figure 2 A schematic diagram of a region extraction image provided in an embodiment of this application;

[0062] Figure 3 A flowchart illustrating an implementation method for extracting the middle artery region to obtain a region extraction image, provided in an embodiment of this application;

[0063] Figure 4 This application provides a flowchart for obtaining the location information of a candidate frame for a high-density sign in the middle cerebral artery;

[0064] Figure 5 For Figure 2 A schematic diagram of the enhanced blood vessel image obtained after extracting blood vessels from the region shown;

[0065] Figure 6 A schematic diagram of the grayscale value distribution curve of a blood vessel enhancement image provided in an embodiment of this application;

[0066] Figure 7 This is a schematic diagram of a blood vessel mask image.

[0067] Figure 8 For based on Figure 7 A schematic diagram of the obtained HMCAS candidate boxes;

[0068] Figure 9 A flowchart illustrating an implementation method for extracting texture features according to an embodiment of this application;

[0069] Figure 10 A schematic diagram of a feature fusion network provided in an embodiment of this application;

[0070] Figure 11 The image shows the HMCAS identified by the technical solution of this application.

[0071] Figure 12 A flowchart illustrating another method for identifying high-density signs of the middle cerebral artery provided in this application embodiment;

[0072] Figure 13 This is a schematic diagram of the structure of a device for identifying high-density signs of the middle cerebral artery, provided in an embodiment of this application. Detailed Implementation

[0073] Manual identification of HMCAS needs to rely on professional experience, and there are problems of low efficiency and easy misjudgment. Although the technology of automatic identification of HMCAS has made breakthroughs in identification efficiency, there are problems of insufficient accuracy and reliability. On the one hand, the existing automatic identification scheme of HMCAS has problems of positioning deviation and poor adaptability to equipment and individual differences, which will seriously affect the identification result; on the other hand, the existing automatic identification scheme of HMCAS is difficult to remove interference in a complex brain lesion environment, which makes it difficult to accurately identify HMCAS and reduces the identification accuracy. For example, factors such as improper head position, brain edema, small nodular lesions, and brain calcification increase the difficulty of identifying HMCAS.

[0074] In combination with the above problems, a solution is provided in the embodiments of the present application, and a method, device and equipment for identifying HMCAS and a storage medium are provided. In the technical scheme of the present application, the middle artery region of the brain CT plain scan image is extracted to obtain a region extraction image; the position information of the HMCAS candidate box is obtained according to the blood vessel enhanced image corresponding to the region extraction image; the texture feature is obtained according to the region extraction image and the position information, and the shape feature is obtained according to the blood vessel enhanced image and the position information; and the HMCAS identification result is obtained according to the texture feature and the shape feature. Through the combined application of the texture feature and the shape feature, the accuracy and reliability of HMCAS identification are greatly improved.

[0075] In order for those skilled in the art to better understand the technical scheme of the present application, the technical scheme of the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0076] Referring to Figure 1 , the figure is a flowchart of a method for identifying HMCAS provided by the embodiments of the present application. As shown in Figure 1 , the method for identifying HMCAS includes:

[0077] S101, extracting the middle artery region of the brain CT plain scan image to obtain a region extraction image.

[0078] In the embodiments of the present application, the brain CT plain scan image refers to the NCCT image commonly used in the medical field. There are various types of regions in the brain CT plain scan image. Considering that HMCAS exists in the middle artery region, in order to facilitate subsequent processing, the middle artery region in the brain CT plain scan image can be extracted by this step to obtain a region extraction image. Figure 2The schematic diagram for extracting the region of the image. The CT image of the brain is a three-dimensional image, each layer of which has a corresponding two-dimensional image, Figure 2 The left and right sides show the extraction effect of the middle cerebral artery region of different layers of the CT image of the brain.

[0079] The following describes an optional implementation of the step. Figure 3 The implementation flowchart for extracting the middle cerebral artery region to obtain a region extraction image. In the implementation, Figure 3 In the provided implementation, the middle cerebral artery region of the CT image of the brain is extracted to obtain a region extraction image, specifically including:

[0080] S1011, removing the region of the non-brain parenchyma in the CT image of the brain by the brain mask to obtain a target brain image.

[0081] The standard brain mask can remove the region irrelevant to the brain parenchyma in the CT image of the brain. For example, the skull part in the image is removed. The image after removing the region of the non-brain parenchyma is taken as the target brain image for subsequent region extraction processing. By removing the region of the non-brain parenchyma, the complexity of subsequent image processing can be simplified, the pertinence of image processing is improved, the accurate extraction of the middle cerebral artery region of the brain is facilitated, and thus the HMCAS recognition accuracy and reliability are improved.

[0082] S1012, registering the target brain image by referring to the brain template image and the middle cerebral artery region label of the brain template image to obtain a candidate middle cerebral artery region label.

[0083] S1013, extracting the region of the target brain image by the candidate middle cerebral artery region label to obtain a region extraction image.

[0084] In the embodiments of the present application, the target brain image can be registered in a rigid and affine registration manner. Rigid and affine registration belongs to the relatively mature technology in the art, which is not described herein. In this implementation, the target brain image is rigidly and affinely registered by referring to the brain template image and the middle cerebral artery region label of the brain template image, which can reduce the interference of individual differences or improper individual positioning on the accuracy of region positioning and extraction, so as to more accurately lock the middle cerebral artery region in the target brain image.

[0085] The process of region extraction by registration mainly involves two links, which correspond to steps S1012 and S1013 described above:

[0086] 1) First, the middle cerebral artery region in the target brain image that can be used to identify HMCAS is obtained by registration, and is identified by a label. In the embodiment of the application, the label is named as a candidate middle cerebral artery region label, which can be in the form of 0 or 1. 1 represents selection, and 0 represents non-selection, so that the image region to be extracted can be distinguished according to the candidate middle cerebral artery region label.

[0087] 2) Since the candidate middle cerebral artery region label has distinguished the image region to be extracted, in the second link, only the middle cerebral artery region can be extracted from the target brain image in which the non-brain parenchymal region has been removed according to the candidate middle cerebral artery region label, to obtain the region extraction image shown in FIG. 2. Figure 2 In the region extraction image, the extracted middle cerebral artery region is identified. In a possible implementation manner, the extracted region is identified in the region extraction image by a color different from other regions; in another possible implementation manner, the gray value of other regions is 0, and only the extracted middle cerebral artery region has a change in the gray value; and in still another possible implementation manner, the outer contour of the middle cerebral artery region in the region extraction image is identified.

[0088] In some possible implementation manners, in order to improve the region extraction effect and reduce the influence of unnecessary interference on the accuracy of the extraction result, denoising processing can also be performed before region extraction. Specifically, the non-brain parenchymal region in the brain CT plain scan image is removed by a brain mask to obtain a target brain image, including: performing denoising processing (for example, performing denoising processing on the image by Gaussian filtering) on the brain CT plain scan image to obtain a processed brain CT plain scan image. The brain mask is extracted from the processed brain CT plain scan image. Finally, the non-brain parenchymal region in the processed brain CT plain scan image is removed by the brain mask to obtain the target brain image. Since the brain mask is extracted from the denoised image, the non-brain parenchymal region can be removed more accurately, the target brain image is more accurate, the noise interference is reduced, and the accuracy of the data basis for subsequent processing operations is improved.

[0089] In the above steps, the extraction of the region is completed, which can be regarded as coarse selection. In order to more accurately identify HMCAS, the position of the blood vessel where HMCAS is likely to appear is finely selected by positioning a candidate box. The following S102 step is used for introduction.

[0090] S102, obtaining the position information of the middle cerebral artery high density sign candidate box according to the blood vessel enhanced image corresponding to the region extraction image.

[0091] The blood vessel enhancement on the region extraction image can reduce the missed detection rate of the small nodular HMCAS. The position information of the HMCAS candidate box is extracted in the following manner: the blood vessels in the image are segmented, and the position of the candidate box is locked according to the connectivity of the segmented image. Figure 4 A flowchart for obtaining the position information of the HMCAS candidate box is provided in the embodiment of the present application. Figure 4 In the implementation shown, the position information of the HMCAS candidate box specifically includes the following steps:

[0092] S1021, the region extraction image is enhanced by filtering to obtain a blood vessel enhanced image corresponding to the region extraction image.

[0093] In the optional implementation, the blood vessel enhancement in the image can be realized based on Frangi filtering. Frangi filtering is a classical blood vessel enhancement and tubular enhancement filtering algorithm, which has excellent mathematical proof and experimental results. The Frangi filtering algorithm is not described in the present application. Figure 5 For the region extraction image Figure 2 The blood vessel enhanced image obtained after the blood vessel enhancement on the region extraction image is shown in the schematic diagram. Figure 5 The left and right sides correspond to the blood vessel enhancement effects of the middle cerebral artery regions in different layers. Figure 2 The blood vessel enhancement effects of the middle cerebral artery regions in different layers on the left and right sides.

[0094] S1022, the blood vessel segmentation threshold is determined according to the distribution of the gray value in the blood vessel enhanced image.

[0095] In the blood vessel enhanced image, the pixel points at different positions may have different gray values. For example, the gray values of some pixel points in the blood vessel enhanced image are large, showing a relatively bright effect; the gray values of other pixel points are small, showing a relatively dark effect. In order to extract the position information of the HMCAS candidate box, the present application proposes to determine the segmentation threshold according to the distribution of the gray value in the blood vessel enhanced image, and the threshold is specifically represented by the gray value. That is, the threshold of the gray value is used to realize the segmentation of the image.

[0096] An example implementation for determining the blood vessel segmentation threshold is provided in the embodiment of the present application. First, the distribution curve of the gray value is calculated according to the gray values of the pixel points other than the pixel points with a gray value of 0 in the blood vessel enhanced image. Figure 6A schematic diagram of a gray value distribution curve of a blood vessel enhanced image is shown. The horizontal axis v of the distribution curve represents the gray value, and the vertical axis f represents the number of pixels. Through the curve graph, the maximum value fmax of the distribution curve on the vertical axis can be determined very conveniently. The value corresponding to fmax on the horizontal axis is called vm, which represents the gray value with the largest number of pixels. A preset coefficient r can be set according to experience or requirements, for example, r is 0.75. The reference number is determined according to the preset coefficient and the maximum value, for example, the reference number fk is obtained by multiplying the preset coefficient r and the maximum value fmax of the distribution curve on the vertical axis, where fk=r*fmax. For the reference number fk, the corresponding horizontal coordinate, that is, the corresponding gray value, can be determined from the curve. In the embodiment of the application, the smallest gray value vb corresponding to fk is taken as the blood vessel segmentation threshold. In actual application, other values of r can be set according to experience or requirements. The value of r is not limited here.

[0097] S1023, processing the blood vessel enhanced image according to the blood vessel segmentation threshold to obtain a blood vessel mask image.

[0098] When v>vb, Cm=1; when v≤vb, Cm=0. The value of Cm represents the gray value of the corresponding pixel point in the blood vessel mask image. V represents the gray value of any pixel in the blood vessel enhanced image. Processing the blood vessel enhanced image according to the blood vessel segmentation threshold is to take a new gray value 1 for the pixel point with a gray value greater than the blood vessel segmentation threshold in the blood vessel enhanced image, and take a new gray value 0 for the pixel point with a gray value less than or equal to the blood vessel segmentation threshold in the blood vessel enhanced image. The image obtained by processing the actual pixel point gray value into a gray value of 0 or 1 is called a blood vessel mask image. Figure 7 A schematic diagram of a blood vessel mask image is shown. Figure 7 The left and right sides are respectively Figure 5 the processing effects of the blood vessel enhanced images shown on the left and right sides.

[0099] S1024, determining the minimum circumscribed rectangle corresponding to each connected region in the blood vessel mask image as a middle cerebral artery high density sign candidate box.

[0100] Through the foregoing step S1023, the screening of the blood vessel graph has been completed. In order to better lock the distribution position of the HMCAS, in this step, the minimum circumscribed rectangle corresponding to each connected region in the blood vessel mask image is taken as the HMCAS candidate box. Figure 8 A schematic diagram of the HMCAS candidate box based on Figure 7 the HMCAS candidate box is obtained. As Figure 8 shown, a plurality of boxes are marked in the figure, which respectively represent different HMCAS candidate boxes.

[0101] S1025, obtaining the position information of the middle cerebral artery high density sign candidate box in the image.

[0102] When the minimum circumscribed rectangle has been determined, the position information of the HMCAS in the image can be obtained according to the positions of the pixels on the minimum circumscribed rectangle.

[0103] The region extraction image, the blood vessel enhanced image corresponding to the region extraction image and the HMCAS candidate box are obtained through S101 and S102 respectively. In order to accurately identify whether the HMCAS exists in the HMCAS candidate box, the present application embodiment proposes to further extract a plurality of types of features as the identification basis for the HMCAS through S103. The acquisition method of different types of features will be described below in combination with S103.

[0104] S103, obtaining texture features according to the region extraction image and the position information, and obtaining shape features according to the blood vessel enhanced image and the position information.

[0105] It is proposed in the present application embodiment to extract texture features and shape features. The texture features are mainly from the region extraction image, and the shape features are mainly from the blood vessel enhanced image. This is because the region extraction image reflects more details of the brain tissue, especially the middle cerebral artery region, especially the texture details; the blood vessel enhanced image reflects relatively accurate shape details.

[0106] In the present application embodiment, the candidate middle cerebral artery high density sign image can be obtained according to the region extraction image and the position information of the HMCAS candidate box, and the candidate blood vessel enhanced image can be obtained according to the blood vessel enhanced image and the position information of the HMCAS candidate box. The position information of the HMCAS candidate box can be used to cut out the candidate middle cerebral artery high density sign image corresponding to the position of the HMCAS candidate box in the region extraction image, and to cut out the candidate blood vessel enhanced image corresponding to the position of the HMCAS candidate box in the blood vessel enhanced image. The candidate middle cerebral artery high density sign image serves as the extraction basis of the texture features, and the candidate blood vessel enhanced image serves as the extraction basis of the shape features.

[0107] S104, obtaining the middle cerebral artery high density sign identification result according to the texture features and the shape features.

[0108] Since the texture features corresponding to the position of the HMCAS candidate box in the region extraction image are extracted in S103, and the shape features corresponding to the position of the HMCAS candidate box in the blood vessel enhanced image are extracted, these types of features can be applied to the HMCAS identification, providing more sufficient criteria.

[0109] In the present application, the position information of the HMCAS candidate box is obtained on the basis of the blood vessel enhanced image corresponding to the region extraction image, so the position of the candidate box has strong position correspondence with the middle cerebral artery region and the related blood vessels. The position can assist in achieving accurate HMCAS identification. In addition, the scheme extracts texture features and shape features based on the position of the candidate box on the region extraction image and the blood vessel enhanced image, respectively. The texture features can reflect the degree of change of the texture of the relevant part in the image, and the shape features can reflect the shape characteristics of the blood vessels in the image. The combination of the two can distinguish the HMCAS from other lesions or influencing factors existing in the image, so as to achieve accurate and reliable identification of HMCAS in a complex brain lesion environment. In addition, the scheme is not simply based on CT value for HMCAS identification, so it has strong adaptability to scanning equipment and individual imaging differences.

[0110] An example implementation of extracting texture features will be introduced below, see Figure 9 . Figure 9 An implementation flowchart of extracting texture features provided by the embodiment of the present application.

[0111] S901, the gray level co-occurrence matrix corresponding to the middle cerebral artery high density sign candidate box in each layer image of the region extraction image is calculated from a plurality of different preset angles.

[0112] The candidate middle cerebral artery high density sign image is extracted from the region extraction image based on the HMCAS candidate box, so the gray level co-occurrence matrix corresponding to the middle cerebral artery high density sign candidate box in each layer image of the region extraction image is essentially the gray level co-occurrence matrix corresponding to each layer candidate middle cerebral artery high density sign image calculated. That is, the gray level co-occurrence matrix is calculated based on the candidate middle cerebral artery high density sign image.

[0113] Considering that the blood vessels can grow in all directions of the three-dimensional image (NCCT) image. In order to more accurately and comprehensively identify HMCAS, a plurality of different preset angles are used to calculate the gray level co-occurrence matrix corresponding to the middle cerebral artery high density sign candidate box in each layer image of the region extraction image in this step. As an example, the plurality of different preset angles include: 0°, 45°, 90° and 135°. Of course, more angles can be set as preset angles based on actual situation and demand, or other angles other than 0°, 45°, 90° and 135° can be taken as preset angles.

[0114] Gray level co-occurrence matrix refers to a commonly used method for describing texture by studying the spatial correlation characteristics of gray levels. Since texture is formed by repeated occurrence of gray level distribution in spatial position, there is a certain gray level relationship between two pixels separated by a certain distance in the image space, i.e. the spatial correlation characteristics of gray levels in the image. Calculating the gray level co-occurrence matrix belongs to the current mature technology, which is not described here. The gray level co-occurrence matrix is represented by g ij , where i represents the layer number of the image, for example, i = 1 represents the first layer of the region extraction image. j = 1, j = 2, j = 3, j = 4 correspond to four different preset angles respectively.

[0115] S902, obtaining the target gray level co-occurrence matrix according to the mean value of the calculated multiple gray level co-occurrence matrices.

[0116] The target gray level co-occurrence matrix is represented by G, G = mean(g ij ), which represents the mean value of the multiple gray level co-occurrence matrices obtained in S901. By calculating the mean value of the multiple gray level co-occurrence matrices, the data for reflecting the texture features calculated at different angles and different layers can be fused.

[0117] S903, constructing the texture feature according to the target gray level co-occurrence matrix.

[0118] The texture feature can be reflected by contrast, cross-correlation, energy and homogeneity, etc. Multiple types of texture features as criteria for identifying HMCAS can also assist in achieving more comprehensive consideration and obtaining more accurate identification results. The contrast, cross-correlation, energy and homogeneity are calculated according to the target gray level co-occurrence matrix as the texture feature. Formulas (1)-(4) show the acquisition methods of contrast, cross-correlation, energy and homogeneity respectively.

[0119]

[0120]

[0121]

[0122]

[0123] In the above formulas, con represents contrast, cor represents cross-correlation, ene represents energy, and hom represents homogeneity. u m , u n , σ m , σ n The four parameter expressions are as follows:

[0124]

[0125]

[0126]

[0127] Based on the above four texture features of contrast, cross-correlation, energy and homogeneity, a texture feature vector ft = [con, cor, ene, hom] can be constructed.

[0128] An example implementation of extracting shape features is introduced as follows.

[0129] In the embodiments of the present application, a tubular feature vector can be obtained according to the gray scale histogram of the candidate blood vessel enhanced image. The specific implementation is as follows:

[0130] s = FI * candNCCT (9)

[0131] v = {s, 10Hu < s < 100Hu} (10)

[0132] w = hist(v) (11)

[0133] fs = w / max(w) (12)

[0134] The value of the blood vessel enhanced image is denoted by FI, which is not in the normal CT gray scale value range, so it needs to be multiplied by candNCCT representing the region extraction image through equation (9) to map it into the normal CT gray scale value range. v represents the data of the non-brain parenchymal gray scale range. w is each value on the gray scale distribution curve. Equation (12) divides w by the maximum value max(w) in w to represent the normalization (data normalization) of the 0-1 numerical interval. fs obtained through the above equation is taken as the tubular feature vector (i.e. shape feature).

[0135] In step S104 of the foregoing embodiments, it is introduced that the HMCAS recognition result can be obtained according to the texture features and shape features. An example implementation of obtaining the HMCAS recognition result is provided as follows. In an optional implementation, the HMCAS recognition result is obtained according to the texture features and shape features, specifically including:

[0136]

[0137] The features of multiple scales of the candidate HMCAS image are acquired. For example, the candidate HMCAS image is expanded to a specified size (for example, 98*96*8) according to different scales, and then features in different fields of view are extracted by 3 times of convolution with a size of 3 and a step of 2, and the features in different fields of view are fused.

[0138] In S103, the texture feature and the shape feature have been obtained. Next, the fused features of multiple scales can be combined with the texture feature ft and the shape feature fs to obtain the combined features Fc. The recognition result of the candidate HMCAS image is obtained according to the combined features. The recognition result has two possibilities: one is that the candidate HMCAS image has HMCAS, and the other is that the candidate HMCAS image does not have HMCAS.

[0139] The above process of recognizing HMCAS can be implemented by a feature fusion network, Figure 10 A schematic diagram of a feature fusion network provided by an embodiment of the present application. Figure 10 F represents the combined features of the shape feature fs and the texture feature ft. Conv represents a convolution network, Add represents tensor addition, and Concat represents tensor splicing. As shown in the feature fusion network, Figure 10 the feature fusion network extracts and fuses the features of multiple scales of the same candidate HMCAS image, fuses multiple different types of features (the shape feature fs and the texture feature ft), and finally outputs the combined features Fc. The combined features Fc are used to recognize and determine whether HMCAS exists in the candidate HMCAS image. Through the fusion of features of multiple scales, the features obtained in different fields of view can be comprehensively considered, and the problem of less and incomplete features in single scale recognition affecting the recognition accuracy can be avoided. Figure 11 An effect diagram of HMCAS recognized by the technical solution of the present application.

[0140] After the identification and judgment of HMCAS in the NCCT image are implemented by the technical solutions introduced in the foregoing embodiments, the present application further proposes that the HMCAS can be further extracted. In an optional implementation manner, the middle cerebral artery high density sign identification method further includes: performing segmentation on the identified candidate middle cerebral artery high density sign image with the middle cerebral artery high density sign, by using a blood vessel mask to obtain a middle cerebral artery high density sign segmentation result. Here, the blood vessel mask can be extracted from the candidate blood vessel enhanced image. In addition, the positions exceeding 100 Hu in the segmentation result of the blood vessel mask can be further removed to obtain a final segmentation result. This is because there can be a part higher than 100 Hu in the individual high density sign candidate box, which is generally a tube during an interventional operation, and there is also a skull that is difficult to remove due to being close to the brain parenchyma. The threshold needs to be removed according to the threshold. Of course, the threshold can also be adjusted in actual application, and is not limited to 100 Hu. By segmenting the HMCAS, the graph of the HMCAS can be output, thereby facilitating the user (for example, a doctor, a consultation expert, and the like) to provide a simple and intuitive image for reference and comparison.

[0141] Figure 12 The flowchart of another middle cerebral artery high density sign identification method provided by the embodiments of the present application is shown. The brain CT plain scan image is extracted to obtain a region extraction image related to the middle cerebral artery region, and a blood vessel enhanced image is processed based on the region extraction image. The blood vessel enhanced image is further processed to obtain a middle cerebral artery high density sign candidate box. The candidate box can obtain a candidate middle cerebral artery high density sign image in combination with the region extraction image, and further obtain a texture feature. The candidate box can obtain a candidate blood vessel enhanced image in combination with the blood vessel enhanced image, and further obtain a shape feature. The features obtained by the multi-scale transformation of the candidate middle cerebral artery high density sign image are fused with the shape feature and the texture feature, and the identification of the HMCAS can be further implemented.

[0142] Based on the middle cerebral artery high density sign identification method provided in the foregoing embodiments, the present application also correspondingly provides a middle cerebral artery high density sign identification device. Figure 13 The structure diagram of a middle cerebral artery high density sign identification device is shown. As Figure 13 The identification device includes:

[0143] The region extraction module 1301 is configured to extract the middle cerebral artery region of the brain CT plain scan image to obtain a region extraction image.

[0144] The candidate box position acquisition module 1302 is configured to obtain the position information of the middle cerebral artery high density sign candidate box according to the blood vessel enhanced image corresponding to the region extraction image.

[0145] The feature extraction module 1303 is configured to obtain a texture feature according to the region extraction image and the position information, and obtain a shape feature according to the blood vessel enhanced image and the position information.

[0146] The recognition module 1304 is configured to obtain a middle cerebral artery hyperdensity sign recognition result according to the texture feature and the shape feature.

[0147] Optionally, the feature extraction module specifically comprises:

[0148] obtaining a candidate middle cerebral artery hyperdensity sign image according to the region extraction image and the position information, and obtaining a candidate blood vessel enhanced image according to the blood vessel enhanced image and the position information;

[0149] extracting a texture feature from the candidate middle cerebral artery hyperdensity sign image, and extracting a shape feature from the candidate blood vessel enhanced image.

[0150] Optionally, the recognition module specifically comprises:

[0151] obtaining a feature of a plurality of scales of the candidate middle cerebral artery hyperdensity sign image;

[0152] merging the feature of the plurality of scales after fusion with the texture feature and the shape feature to obtain a merged feature;

[0153] obtaining a recognition result of the candidate middle cerebral artery hyperdensity sign image according to the merged feature, the recognition result being having a middle cerebral artery hyperdensity sign or not having a middle cerebral artery hyperdensity sign.

[0154] Optionally, the middle cerebral artery hyperdensity sign recognition device further comprises:

[0155] The segmentation module is configured to segment a candidate middle cerebral artery hyperdensity sign image having a middle cerebral artery hyperdensity sign recognized to obtain a middle cerebral artery hyperdensity sign segmentation result by using a blood vessel mask.

[0156] Optionally, the feature extraction module specifically comprises:

[0157] The matrix obtaining unit is configured to calculate a gray level co-occurrence matrix corresponding to the middle cerebral artery hyperdensity sign candidate box in each layer image of the region extraction image from a plurality of different preset angles respectively;

[0158] The mean value calculation unit is configured to obtain a target gray level co-occurrence matrix by calculating a mean value according to the plurality of calculated gray level co-occurrence matrices;

[0159] The texture feature construction unit is configured to construct a texture feature according to the target gray level co-occurrence matrix.

[0160] Optionally, the texture feature construction unit is specifically configured to:

[0161] According to the target gray level co-occurrence matrix, contrast, cross-correlation, energy and homogeneity are calculated as texture features.

[0162] Optionally, the region extraction module specifically comprises:

[0163] The region elimination unit is configured to eliminate the region of non-brain parenchyma in the brain CT plain scan image through the brain mask to obtain a target brain image.

[0164] The registration unit is configured to register the target brain image through the reference brain template image and the middle artery region label in the reference brain template image to obtain a candidate middle artery region label.

[0165] The region extraction unit is configured to extract the region of the target brain image through the candidate middle artery region label to obtain a region extraction image.

[0166] Optionally, the region elimination unit is specifically configured to:

[0167] The brain CT plain scan image is denoised to obtain a processed brain CT plain scan image.

[0168] The brain mask is extracted from the processed brain CT plain scan image.

[0169] The region of non-brain parenchyma in the processed brain CT plain scan image is eliminated through the brain mask to obtain a target brain image.

[0170] Optionally, the candidate box position acquisition module specifically comprises:

[0171] The blood vessel enhancement unit is configured to enhance the blood vessels of the region extraction image in a filtering manner to obtain a blood vessel enhancement image corresponding to the region extraction image.

[0172] The threshold determination unit is configured to determine a blood vessel segmentation threshold according to the distribution of the gray value in the blood vessel enhancement image.

[0173] The segmentation unit is configured to process the blood vessel enhancement image according to the blood vessel segmentation threshold to obtain a blood vessel mask image.

[0174] The candidate box determination unit is configured to determine the minimum circumscribed rectangle corresponding to each connected region in the blood vessel mask image as a middle cerebral artery high density sign candidate box.

[0175] The position acquisition unit is configured to acquire the position information of the middle cerebral artery high density sign candidate box in the image.

[0176] Optionally, the threshold determination unit is specifically configured to:

[0177] According to the gray value of the pixel points other than the pixel points with the gray value of 0 in the blood vessel enhanced image, a distribution curve of the gray value is calculated; the horizontal axis of the distribution curve represents the gray value, and the vertical axis represents the number of pixel points;

[0178] The maximum value of the distribution curve on the vertical axis is determined;

[0179] A reference number is determined according to a preset coefficient and the maximum value;

[0180] A minimum gray value corresponding to the reference number is determined from the distribution curve as a blood vessel segmentation threshold.

[0181] The application further provides a middle cerebral artery hyperdensity sign identification device. The identification device comprises:

[0182] A memory having a computer program stored thereon.

[0183] A processor configured to execute the computer program in the memory to implement some or all steps of the middle cerebral artery hyperdensity sign identification method described in the foregoing embodiments.

[0184] The application further provides a computer readable storage medium having a computer program stored thereon. The program, when executed by a processor, implements some or all steps of the middle cerebral artery hyperdensity sign identification method described in the foregoing embodiments.

[0185] It should be noted that each of the embodiments in the present specification is described in a progressive manner, and the same or similar parts of each embodiment can be understood by mutual reference. Each embodiment focuses on the differences from other embodiments. In particular, the device and equipment embodiments are described more simply because they are basically similar to the method embodiments, and the relevant parts can be understood by referring to the part of the method embodiments. The device and equipment embodiments described above are only illustrative, and the units described as separate components can be or can not be physically separated, and the components indicated as units can be or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Some or all modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement it without creative labor.

[0186] The above describes only one specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any changes or replacements within the technical range disclosed in the present application can be easily thought of by those skilled in the art, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of identifying a high-density sign of middle cerebral artery, characterized by, The method comprises the following steps: extracting a middle cerebral artery region of a brain CT plain scan image to obtain a region extraction image; obtaining position information of a middle cerebral artery high density sign candidate box according to a blood vessel enhanced image corresponding to the region extraction image; obtaining texture features according to the region extraction image and the position information, and obtaining shape features according to the blood vessel enhanced image and the position information; obtaining a middle cerebral artery high density sign recognition result according to the texture features and the shape features; the step of obtaining position information of a middle cerebral artery high density sign candidate box according to a blood vessel enhanced image corresponding to the region extraction image specifically comprises the following steps: performing blood vessel enhancement on the region extraction image by using a filtering mode to obtain the blood vessel enhanced image corresponding to the region extraction image; determining a blood vessel segmentation threshold according to the distribution of gray values in the blood vessel enhanced image; processing the blood vessel enhanced image according to the blood vessel segmentation threshold to obtain a blood vessel mask image; determining the minimum circumscribed rectangle corresponding to each connected region in the blood vessel mask image as the middle cerebral artery high density sign candidate box; obtaining the position information of the middle cerebral artery high density sign candidate box in the image.

2. The method of claim 1, wherein, the step of obtaining texture features according to the region extraction image and the position information, and obtaining shape features according to the blood vessel enhanced image and the position information specifically comprises the following steps: obtaining a candidate middle cerebral artery high density sign image according to the region extraction image and the position information, and obtaining a candidate blood vessel enhanced image according to the blood vessel enhanced image and the position information; extracting texture features from the candidate middle cerebral artery high density sign image, and extracting shape features from the candidate blood vessel enhanced image.

3. The method of claim 2, wherein, the step of obtaining a middle cerebral artery high density sign recognition result according to the texture features and the shape features specifically comprises the following steps: obtaining features of multiple scales of the candidate middle cerebral artery high density sign image; merging the features of multiple scales after fusion with the texture features and the shape features to obtain merged features; obtaining a recognition result of the candidate middle cerebral artery high density sign image according to the merged features, wherein the recognition result is a middle cerebral artery high density sign or no middle cerebral artery high density sign.

4. The method of claim 3, wherein, It also comprises the following steps: performing segmentation on the candidate middle cerebral artery high density sign image with a blood vessel mask to obtain a middle cerebral artery high density sign segmentation result.

5. The method according to any one of claims 1 to 4, characterized in that, the step of obtaining texture features according to the region extraction image and the position information specifically comprises the following steps: calculating a gray level co-occurrence matrix corresponding to the middle cerebral artery high density sign candidate box in each layer image of the region extraction image from multiple different preset angles respectively; obtaining a target gray level co-occurrence matrix by averaging the multiple calculated gray level co-occurrence matrices; constructing texture features according to the target gray level co-occurrence matrix.

6. The method of claim 5, wherein, the step of constructing texture features according to the target gray level co-occurrence matrix comprises the following steps: calculating contrast, cross-correlation, energy and homogeneity as texture features according to the target gray level co-occurrence matrix.

7. The method according to any one of claims 1 to 4, characterized in that, the step of extracting a middle cerebral artery region of a brain CT plain scan image to obtain a region extraction image specifically comprises the following steps: The target brain image is obtained by removing the non-brain parenchymal region in the brain CT scan image through the brain mask; The candidate middle cerebral artery region label is obtained by registering the target brain image by referring to the brain template image and the middle cerebral artery region label of the brain template image; The region extraction image is obtained by region extraction of the target brain image through the candidate middle cerebral artery region label.

8. The method of claim 7, wherein, The target brain image is obtained by removing the non-brain parenchymal region in the brain CT scan image through the brain mask, specifically comprising: The brain CT scan image is denoised to obtain a processed brain CT scan image; The brain mask is extracted from the processed brain CT scan image; The target brain image is obtained by removing the non-brain parenchymal region in the processed brain CT scan image through the brain mask.

9. The method of claim 1, wherein, The blood vessel segmentation threshold is determined according to the distribution of the gray value in the blood vessel enhanced image, specifically comprising: The distribution curve of the gray value is calculated according to the gray value of the remaining pixel points except the pixel points with a gray value of 0 in the blood vessel enhanced image; the horizontal axis of the distribution curve represents the gray value, and the vertical axis represents the number of pixel points; The maximum value of the distribution curve on the vertical axis is determined; The reference number is determined according to the preset coefficient and the maximum value; The smallest gray value corresponding to the reference number is determined from the distribution curve as the blood vessel segmentation threshold.

10. A device for identifying a hyperdense middle cerebral artery sign, characterized in that Comprise: The region extraction module is used for extracting the middle cerebral artery region of the brain CT scan image to obtain a region extraction image; The candidate box position acquisition module is used for obtaining the position information of the middle cerebral artery high density sign candidate box according to the blood vessel enhanced image corresponding to the region extraction image; The feature extraction module is used for obtaining the texture feature according to the region extraction image and the position information, and obtaining the shape feature according to the blood vessel enhanced image and the position information; The recognition module is used for obtaining the middle cerebral artery high density sign recognition result according to the texture feature and the shape feature; The candidate box position acquisition module specifically comprises: The blood vessel enhancement unit is used for performing blood vessel enhancement on the region extraction image in a filtering manner to obtain the blood vessel enhanced image corresponding to the region extraction image; The threshold determination unit is used for determining the blood vessel segmentation threshold according to the distribution of the gray value in the blood vessel enhanced image; The segmentation unit is used for processing the blood vessel enhanced image according to the blood vessel segmentation threshold to obtain a blood vessel mask image; The candidate box determination unit is used for determining the minimum circumscribed rectangle corresponding to each connected region in the blood vessel mask image as the middle cerebral artery high density sign candidate box; The position acquisition unit is used for acquiring the position information of the middle cerebral artery high density sign candidate box in the image.

11. A device for identifying a hyperdense middle cerebral artery sign, characterized in that Comprise: The memory has a computer program stored thereon; The processor is used for executing the computer program in the memory to realize the steps of the middle cerebral artery high density sign recognition method in any one of claims 1-9.

12. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the steps of the middle cerebral artery high density sign recognition method in any one of claims 1-9.

Images