A region detection method, device, apparatus and storage medium

Abstract

Embodiments of the present application disclose a region detection method, device and equipment and a storage medium. The method comprises: clustering each gray pixel point in a medical image according to the pixel gray of each gray pixel point, and taking a region in which each gray pixel point belonging to the same category is located as a candidate region; generating a gradient image corresponding to the medical image according to each pixel gray, and determining a gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image; and detecting a target region from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge. The technical solution of the embodiments of the present application solves the problem of low detection accuracy of the target region caused by various reasons based on clustering, and realizes the effect of accurate detection of the target region by comparing the complete region boundary of the candidate region obtained thereby with the gradient edge of the region boundary which is more likely to be the target region.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of medical image processing, and in particular to a region detection method, device, equipment and storage medium. BACKGROUND

[0002] In X-ray images, there are often low gray value low gray regions and / or high gray value high gray regions; the low gray region can come from implants (such as artificial joints, stents, pacemakers, steel plates, screws, etc.) in the body of a subject, fixation devices (such as external fixation devices) used to fix the subject during a surgical procedure, positioning devices (such as positioning needles, clips, etc.) used to position lesions, imaging devices (such as needle holding devices in breast puncture images, etc.) in certain medical devices, etc., which can also be referred to as high attenuation regions; the high gray region can come from regions in the X-ray image that do not pass through the body, which can also be referred to as direct exposure regions.

[0003] Generally, the high attenuation region and the direct exposure region in the X-ray image are not the regions that the doctor needs to pay attention to, and they will interfere with the imaging effect of normal human tissues. Specifically, through feedback from clinical problems and phenomena, it is shown that if there are high attenuation regions and / or direct exposure regions in the X-ray image, this will directly affect the display effect of normal human tissues under the default window width and window level, so it is necessary to detect the high attenuation region and / or the direct exposure region from the X-ray image.

[0004] It should be noted that, due to the undefined gray difference between the high attenuation region and the direct exposure region in the X-ray image and the normal human tissue, and the gray variation within the high attenuation region and the direct exposure region, plus the influence of image noise and other factors, it is very difficult to accurately detect the high attenuation region and / or the direct exposure region from the X-ray image. SUMMARY

[0005] Embodiments of the present application provide a region detection method, device, equipment and storage medium to achieve the effect of accurate detection of a target region in a medical image.

[0006] In a first aspect, embodiments of the present application provide a region detection method, which can include:

[0007] According to the pixel gray of each gray pixel point in the medical image, the gray pixel points are clustered, and the region where each gray pixel point belonging to the same category is located is taken as a candidate region; a gradient image corresponding to the medical image is generated according to each pixel gray, and a gradient edge within the gradient image is determined according to the pixel gradient of each gradient pixel point in the gradient image; and a target region is detected from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge.

[0008] Optionally, the clustering of the gray-scale pixel points according to the pixel gray-scale of the gray-scale pixel points in the medical image can include: sorting the pixel gray-scale of the gray-scale pixel points in the medical image, and determining a gray-scale category split point in the gray-scale sorting result based on a pre-set number of categories, taking the gray-scale category split point as an initial clustering center; and clustering the gray-scale pixel points based on the clustering center and the pixel gray-scale of the gray-scale pixel points.

[0009] On this basis, the clustering of the gray-scale pixel points based on the clustering center and the pixel gray-scale of the gray-scale pixel points can include: for each gray-scale pixel point, determining a gray-scale distance between the pixel gray-scale of the gray-scale pixel point and each clustering center, and clustering the gray-scale pixel point into a category in which the clustering center corresponding to the smallest gray-scale distance of the gray-scale pixel point is located; determining a gray-scale distortion according to the smallest gray-scale distance corresponding to each gray-scale pixel point, and judging whether the gray-scale distortion satisfies a pre-set clustering end condition; if not, for each category, re-determining the clustering center of the category according to the pixel gray-scale of the gray-scale pixel points belonging to the category after clustering; repeating the step of determining the gray-scale distance between the pixel gray-scale of the gray-scale pixel point and each clustering center until the gray-scale distortion satisfies the clustering end condition, and the clustering ends.

[0010] Optionally, the determination of the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image can include: selecting a gray-scale category region corresponding to the region attribute of the target region to be detected from each candidate region according to the region gray-scale of each candidate region; and determining the gradient edge in the gradient image according to the pixel gradient of the gradient pixel point corresponding to each region pixel point in the gray-scale category region in the gradient image.

[0011] Optionally, the determination of the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image can include: sorting each gradient pixel point according to the pixel gradient of each gradient pixel point in the gradient image, and selecting each gradient pixel point according to the sorting position of each gradient pixel point in the sorting result to generate a binary image corresponding to the gradient image; and taking a binary edge formed by each binary pixel point in the binary image as the gradient edge of the gradient image.

[0012] Optionally, the detecting the target region from the candidate regions according to the coupling degree between the region boundary of each candidate region and the gradient edge can include: obtaining a first similarity of each current pixel point in a current boundary of the current region and each edge pixel point in the gradient edge in pixel position, and a second similarity of each current pixel point and each target pixel point in a target boundary of a candidate region detected as the target region in pixel position; and determining whether the current region is the target region according to the first similarity or the first similarity and the second similarity to realize the detection of the target region.

[0013] Optionally, the detecting the target region from the candidate regions according to the coupling degree between the region boundary of each candidate region and the gradient edge can include: determining a region detection order of the candidate regions according to a region attribute of the target region to be detected; screening a current region from the candidate regions according to the region detection order; determining whether the current region is the target region according to the coupling degree between the region boundary of the current region and the gradient edge, and realizing the detection of the target region according to a result of the determination; taking a next region of the current region in the region detection order as the current region, and repeating the step of determining whether the current region is the target region according to the coupling degree between the region boundary of the current region and the gradient edge until the candidate regions and / or the gradient edge meet a pre-set determination end condition.

[0014] In a second aspect, an embodiment of the present application further provides a region detection device, which can include:

[0015] a candidate region determination module configured to cluster each gray-scale pixel point in the medical image according to a pixel gray scale of the gray-scale pixel point, and take a region in which each gray-scale pixel point belonging to a same category is located as a candidate region;

[0016] a gradient edge determination module configured to generate a gradient image corresponding to the medical image according to each pixel gray scale, and determine a gradient edge in the gradient image according to a pixel gradient of each gradient pixel point in the gradient image;

[0017] a target region detection module configured to detect a target region from the candidate regions according to a coupling degree between a region boundary of each candidate region and the gradient edge.

[0018] In a third aspect, an embodiment of the present application further provides a region detection equipment, which can include:

[0019] one or more processors;

[0020] a memory configured to store one or more programs;

[0021] When one or more programs are executed by one or more processors, the one or more processors implement the region detection method provided by any embodiment of the present application.

[0022] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the region detection method provided by any embodiment of the present application.

[0023] The technical scheme of the embodiment of the present application clusters each gray pixel point according to the pixel gray of each gray pixel point in the medical image, takes the region where each gray pixel point belonging to the same category is located as a candidate region, generates a gradient image corresponding to the medical image according to each pixel gray, and determines a gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image. The target region is detected from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge. The above technical scheme solves the problem of low detection accuracy of the target region caused by the gray difference between the target region and the human tissue, image noise, and gray transition in the target region, and realizes the effect of accurate detection of the target region by comparing the complete region boundary of the candidate region obtained in this way with the region boundary of the gradient edge which is more likely to be the target region. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flowchart of a region detection method in the embodiment one of the present application;

[0025] Figure 2 is a flowchart of a region detection method in the embodiment two of the present application;

[0026] Figure 3 is a flowchart of a region detection method in the embodiment three of the present application;

[0027] Figure 4a is a schematic diagram of an optional example of a region detection method in the embodiment three of the present application;

[0028] Figure 4b is a schematic diagram of an optional example of a region detection method in the embodiment three of the present application;

[0029] Figure 5 is a structural block diagram of a region detection device in the embodiment four of the present application;

[0030] Figure 6 is a structural schematic diagram of a region detection device in the embodiment five of the present application. DETAILED DESCRIPTION

[0031] The application will be described in further detail below with reference to the drawings and embodiments. It is to be understood that the specific embodiments described herein are intended to be illustrative only and not limiting of the application. It is also to be understood that the terminology used herein is for the purpose of describing the specific embodiments only and is not intended to be limiting.

[0032] The detection scheme of the target region in the prior art is usually realized based on a gray threshold, region growing and the like, but since the dose is set according to the body state of the examinee in the acquisition process of the X-ray image, and the features of different high-attenuation regions and / or direct exposure regions are unknown, it is difficult to detect the high-attenuation regions and / or direct exposure regions by setting an accurate gray threshold in the detection algorithm based on the gray threshold; in addition, affected by image noise and the like, it is difficult to accurately set the growth criterion in the detection algorithm based on region growing, at this time, over-segmentation or under-segmentation is prone to occur, and these schemes are difficult to meet the actual needs in clinical application.

[0033] Embodiment one

[0034] Figure 1 is a flowchart of a region detection method provided in the embodiment one of the application. The embodiment can be applicable to the case of detecting a target region from a medical image, and is especially applicable to the case of detecting a target region from a medical image based on the coupling degree between region boundaries and gradient edges. The method can be executed by a region detection device provided in the embodiment of the application, the device can be realized in the form of software and / or hardware, and the device can be integrated on a region detection equipment, which can be a terminal or a server.

[0035] Referring to Figure 1 , the method of the embodiment of the application specifically includes the following steps:

[0036] S110, clustering each gray pixel point according to the pixel gray of each gray pixel point in the medical image, and taking the region where each gray pixel point belonging to the same category is located as a candidate region.

[0037] The medical image can be an image obtained by image acquisition on human tissues based on medical imaging technology, such as an X-ray image, a Computed Tomography (CT) image, a Digital Radiography (DR) image, a B-ultrasound image and the like.

[0038] For the imaging object in the medical imaging process, it can include human tissue and the rest of the imaging object other than human tissue. Generally, the human tissue and the rest of the imaging object have different degrees of attenuation of the rays involved in the medical imaging. The higher the degree of attenuation, the lower the pixel gray scale in the medical image; the lower the degree of attenuation, the higher the pixel gray scale in the medical image. Therefore, the imaging area of the rest of the imaging object in the medical image can be taken as a target area to be detected. The target area can be a low gray scale area (i.e., a high attenuation area) with a pixel gray scale obviously lower than the human tissue, a high gray scale area (i.e., a direct exposure area) with a pixel gray scale obviously higher than the human tissue, or a combination of the two, etc.

[0039] The medical image is a gray scale image, the gray scale pixel point is a pixel point in the medical image, and the pixel gray scale is a pixel value (i.e., a gray scale value) of the gray scale pixel point. The gray scale pixel points are clustered according to the pixel gray scales of the gray scale pixel points, i.e., the gray scale pixel points with similar pixel gray scales are classified into the same category, and the gray scale pixel points with large differences in the pixel gray scales are classified into different categories. The clustering can be implemented in various ways, such as the LBG clustering algorithm, the k-means clustering algorithm, etc., which are not limited herein. After clustering the gray scale pixel points, each gray scale pixel point can be clustered into a corresponding category. At this time, the area where the gray scale pixel points belonging to the same category are located can be taken as a candidate area. In other words, the gray scale pixel points belonging to the same candidate area are the gray scale pixel points belonging to the same category, and these gray scale pixel points have strong similarity in the pixel gray scale. It should be noted that the number of the candidate areas composed of the gray scale pixel points belonging to the same category can be one, two or more, which are not limited herein.

[0040] On this basis, if the gray scale pixel points are directly clustered, the gray scale pixel points belonging to the same category may not form a candidate area due to being scattered points, or there may be holes in the candidate area formed (i.e., most of the gray scale pixel points in a certain area belong to the same category, and a small number of gray scale pixel points do not belong to the category), etc. On this basis, in order to ensure that a complete candidate area can be obtained after clustering, the following optional processing methods can be performed: noise reduction, smoothing, etc. on the medical image before clustering to remove noise in the medical image, or region hole filling, etc. on the candidate area obtained after clustering, which are not limited herein.

[0041] In S120, a gradient image corresponding to the medical image is generated according to the pixel gray scales, and a gradient edge in the gradient image is determined according to the pixel gradient of each gradient pixel point in the gradient image.

[0042] The gradient image is an image corresponding to the medical image generated by gradient operation on the pixel gray scale of each gray scale pixel point in the medical image. The gradient operation process can be understood as the calculation process of gradient information involved in the edge extraction process of the medical image, such as the calculation process of gradient information in the edge extraction process based on local variance, sobel, prewitt, canny, etc. For example, to calculate the local variance in the medical image to generate a gradient image, a sliding window with a preset window size of W*H is obtained, the sliding window is constantly moved, the variance of the pixel gray scale of all gray scale pixel points in the sliding window after each movement is calculated in turn, and the average value of all variances is taken as the pixel gradient (i.e. pixel value) of the gradient pixel point corresponding to a gray scale pixel point in the sliding window. The gradient pixel point is a pixel point in the generated gradient image, that is, each gradient pixel point in the gradient image has a unique gray scale pixel point corresponding to it in the pixel position, thereby generating a gradient image corresponding to the medical image.

[0043] The gradient edge is an edge composed of multiple gradient pixel points in the gradient image. The gradient pixel points on the gradient edge can be referred to as edge pixel points. The edge pixel points can be gradient pixel points with large pixel gradients. The significance of setting the gradient edge lies in that there is a strong gray scale difference between the pixel gray scales of the gray scale pixel points on both sides of the region boundary of the target region. The region boundary is the boundary of the target region, which means that the pixel gradient of the gradient pixel point corresponding to the gray scale pixel point on the region boundary is relatively large, that is, the gradient edge is likely to be the region boundary of the target region. Therefore, the candidate region where the region boundary similar to the gradient edge is located can be taken as the target region in the subsequent process.

[0044] On this basis, there are various implementation manners for determining the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image. For example, the gradient edge can be obtained by performing binaryzation processing on the gradient image to obtain a binary image, and the edge composed of each binary pixel point with a pixel value of 1 in the binary image is taken as the gradient edge. For another example, taking the high-attenuation region as an example, the edge pixel points on the gradient edge corresponding to the high-attenuation region are mostly corresponding to the gray scale pixel points in the candidate region with small region gray scale. The region gray scale can represent the overall level of the pixel gray scale of each gray scale pixel point in the candidate region. Therefore, the gray scale category region can be first screened from each candidate region according to the region attribute of the target region to be detected. The region attribute can represent the overall level of the pixel gray scale of each gray scale pixel point in the target region, which can reflect whether the target region is a direct exposure region or a high-attenuation region. Then, the edge pixel points are screened from the gradient pixel points respectively corresponding to each region pixel point in the gray scale category region, and the edge composed of each edge pixel point is taken as the gradient edge. And so on, which is not limited herein.

[0045] S130, detecting a target region from each candidate region according to a coupling degree between the region boundary and the gradient edge of each candidate region.

[0046] The coupling degree can be a similarity between the region boundary and the gradient edge of each candidate region, and specifically, can be a similarity in pixel position between each boundary pixel point on the region boundary and each edge pixel point on the gradient edge. The coupling degree can be calculated in various ways, such as obtaining a first number of boundary pixel points on the region boundary, and a second number of boundary pixel points that can be the same or similar in pixel position to the edge pixel point, and determining the coupling degree according to a number ratio between the second number and the first number. Further, the candidate region corresponding to the region boundary with a higher coupling degree can be taken as the target region. Of course, in addition to considering the coupling degree between the region boundary and the gradient edge, it is also possible that a certain region boundary has a certain coupling degree with both the gradient edge and the target boundary of the candidate region that has been detected as the target region, so such a candidate region can also be the target region. Therefore, the coupling degree between the region boundary and the gradient edge, and the coupling degree between the region boundary and the target boundary can be used together to determine whether the candidate region corresponding to the region boundary is the target region. And so on, which is not specifically limited here.

[0047] It should be noted that the reason for jointly selecting the target region according to the gradient edge and the region boundary, rather than directly selecting the target region according to the gradient edge, is that there can be an overlap between the high-attenuation imaging object and the human tissue, which makes the gradient edge on the high-attenuation region of the high-attenuation imaging object in the medical image have a breakpoint, which means that the target region cannot be detected by region filling on the gradient edge with the breakpoint. Accordingly, the above region detection method can obtain complete candidate regions based on the clustering algorithm, and the region boundary and the gradient edge do not need to be completely coincident, thereby achieving the effect of still being able to detect complete target regions in the case of a breakpoint in the gradient edge.

[0048] The technical scheme of the embodiment of the present application clusters each gray pixel point in a medical image according to the pixel gray of each gray pixel point, takes the region where each gray pixel point belonging to the same category is located as a candidate region, generates a gradient image corresponding to the medical image according to each pixel gray, and determines the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image. The target region is detected from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge. The above technical scheme solves the problem of low detection accuracy of the target region caused by the gray difference between the target region and the human tissue, image noise, and gray transition in the target region, and realizes the effect of accurate detection of the target region by comparing the complete region boundary of the candidate region with the gradient edge which is the region boundary of the target region with a large probability.

[0049] An optional technical scheme determines the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image, which can specifically include: screening the gray category region corresponding to the region attribute of the target region to be detected from each candidate region according to the region gray of each candidate region; and determining the gradient edge in the gradient image according to the pixel gradient of the gradient pixel point corresponding to each region pixel point in the gray category region in the gradient image. In this way, considering that the target region can be a high attenuation region or a direct exposure region, the high attenuation region is usually detected from the candidate region with a smaller region gray, and the direct exposure region is usually detected from the candidate region with a larger region gray, so the gray category region corresponding to the region attribute of the target region to be detected can be screened from each candidate region first, that is, the gray category region is the candidate region similar to the target region in the overall level of the pixel gray; and then the gradient edge in the gradient image is determined according to the pixel gradient of the gradient pixel point corresponding to each region pixel point in the gray category region in the gradient image, for example, considering that the pixel gradient of the gradient pixel point corresponding to the gray pixel point on the region boundary of the target region is relatively large, so the edge formed by each gradient pixel point with a relatively large pixel gradient in the gray category region can be taken as the gradient edge. The above technical scheme determines the gradient edge on the pixel gradient corresponding to the region attribute similar to the target region, thereby improving the determination speed and accuracy of the gradient edge; and the determination of the gradient edge means that the detection of the direct exposure region can be realized on the high gray candidate region, and the detection of the high attenuation region can be realized on the low gray candidate region. Since the high gray candidate region and the low gray candidate region are both adaptive clustering results not affected by the dose, the above technical scheme has good adaptability to medical images of different doses and target regions of different types, thereby realizing the effect of accurate detection of the target region.

[0050] An optional technical solution considers that the pixel gradient of the gradient pixel corresponding to the gray pixel on the region boundary of the target region is generally large. The gradient edges in the gradient image are determined according to the pixel gradient of each gradient pixel in the gradient image. Specifically, the pixel gradient of each gradient pixel in the gradient image is taken as the basis to sort the gradient pixels, and the gradient pixels are filtered according to the sorting position of each gradient pixel in the sorting result to generate a binary image corresponding to the gradient image, that is, the gradient pixels with large pixel gradient in the gradient image are retained, and the gradient pixels with small pixel gradient in the gradient image are discarded. The sorting position can reflect the relative size of the pixel gradient of a certain gradient pixel among all gradient pixels. The binary edge formed by the binary pixels in the binary image is taken as the gradient edge of the gradient image. Since the pixel value of the binary pixel is 1 or 0, the binary edge formed by the binary pixels with a pixel value of 1 can be taken as the gradient edge. The above technical solution sorts the gradient pixels according to the pixel gradient, quickly and accurately filters the gradient pixels belonging to the gradient edge from the gradient pixels, and then obtains the gradient edge formed by these gradient pixels, thereby achieving the effect of accurately determining the gradient edge.

[0051] In order to better understand the specific implementation process of the above gradient edge determination, the following will be exemplarily described in combination with specific examples. Exemplarily, generally, after clustering each gray pixel, the region gray of each candidate region can be obtained. The region gray of the candidate regions belonging to the same category is similar, and the difference of the region gray of the candidate regions belonging to different categories is large. Assuming that the number of preset categories is N, 1 / 2 of which is taken as a low gray category, and the other 1 / 2 is taken as a high gray category, the category of each candidate region can be determined according to the numerical value of the region gray of each candidate region, thereby obtaining the low gray candidate region belonging to the low gray category and the high gray candidate region belonging to the high gray category. When the target region to be detected is a high attenuation region, the low gray candidate region at this time is the gray category region described above. The pixel gradient of the gradient pixel corresponding to each region pixel in the gray category region in the gradient image is taken as the basis to sort these gradient pixels, and a certain proportion of edge pixels is extracted from the sorting result to generate a binary image corresponding to the gradient sub-image, that is, the gradient sub-image is the part of the gradient image corresponding to the gray category region. In this way, the target region is detected from the low gray candidate region subsequently, which improves the detection speed and efficiency of the target region. Of course, when the target region to be detected includes a direct exposure region, the high gray candidate region can be taken as the gray category region to perform the corresponding steps, which is similar to the execution process of the high attenuation region, and will not be described here.

[0052] Embodiment Two

[0053] Figure 2 is a flowchart of a region detection method provided in Embodiment Two of the present application. This embodiment is optimized based on the technical solutions in the above embodiments. In this embodiment, optionally, the gray-scale pixels in the medical image are clustered according to the pixel gray-scale of each gray-scale pixel, which can specifically include: sorting the pixel gray-scale of each gray-scale pixel in the medical image, and determining a gray-scale category split point in the gray-scale sorting result based on a pre-set category number; taking the gray-scale category split point as an initial clustering center, and clustering each gray-scale pixel based on the clustering center and the pixel gray-scale of each gray-scale pixel. The explanation of the same or corresponding terms as in the above embodiments is not repeated here.

[0054] Referring to Figure 2 , the method of this embodiment can specifically include the following steps:

[0055] S210, sorting the pixel gray-scale of each gray-scale pixel in the medical image, and determining a gray-scale category split point in the gray-scale sorting result based on a pre-set category number.

[0056] In the process of clustering each gray-scale pixel, the gray-scale category split point as the initial clustering center will have certain influence on the clustering speed and clustering accuracy. Therefore, in order to improve the clustering effect, the pixel gray-scale of each gray-scale pixel in the medical image can be sorted first, and then a gray-scale category split point is determined in the gray-scale sorting result based on a pre-set category number. For example, assuming that a certain medical image includes 900 gray-scale pixels, and the category number is 90, then the pixel gray-scale ranked at the 10th, 20th, …, 900th can be taken as the gray-scale category split point. In actual application, optionally, since the region boundary obtained by the gradient edge is used to delineate the region boundary of the clustering to detect the target region, in the case where the category number is very small, the region boundary is very likely to be greater than the gradient edge, which means that the region boundary cannot be delineated based on the gradient edge, i.e., the candidate region corresponding to the target region obtained by the clustering can be an under-segmentation result of the target region, which facilitates the subsequent traversal of the region boundary of each candidate region to make it gradually approach the gradient edge, so the category number can be a larger value.

[0057] S220, taking the gray-scale category split point as an initial clustering center, clustering each gray-scale pixel based on the clustering center and the pixel gray-scale of each gray-scale pixel, and taking the region where each gray-scale pixel belonging to the same category is located as a candidate region.

[0058] The gray scale category split point can be used as an initial clustering center, and each gray scale pixel point is clustered based on the clustering center and the pixel gray scale of each gray scale pixel point, for example, for each gray scale pixel point, the gray scale distance between the gray scale pixel point and each clustering center is compared, and the category in which the clustering center corresponding to the smallest gray scale distance is located is used as the clustering result of the gray scale pixel point, that is, the gray scale pixel point is classified into the category in which the clustering center corresponding to the smallest gray scale distance is located, thereby achieving the effect of accurate clustering of each gray scale pixel point.

[0059] On this basis, in order to further improve the clustering accuracy, the following scheme can be used for clustering: for each gray scale pixel point, the pixel gray scale of the gray scale pixel point and the gray scale distance between each clustering center are determined, and the gray scale pixel point is clustered into the category in which the clustering center corresponding to the smallest gray scale distance is located; the gray scale distortion is determined according to the smallest gray scale distance corresponding to each gray scale pixel point, and it is judged whether the gray scale distortion satisfies a pre-set clustering end condition, which can be whether the gray scale distortion is less than a pre-set threshold, whether the absolute value of the difference between the gray scale distortion of this iteration and the gray scale distortion of the last iteration is less than a relative error threshold, etc.; if not, for each category, the clustering center of the category is re-determined according to the pixel gray scale of each gray scale pixel point belonging to the category after clustering, and the step of determining the pixel gray scale of the gray scale pixel point and the gray scale distance between each clustering center is repeated until the gray scale distortion satisfies the clustering end condition, and the clustering ends.

[0060] S230, generating a gradient image corresponding to the medical image according to each pixel gray scale, and determining a gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image.

[0061] S240, detecting a target region from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge.

[0062] The technical scheme of the embodiment of the application determines the gray scale category split point (i.e. the initial clustering center) in the sorting result of the pixel gray scale of each gray scale pixel point by using a pre-set number of categories, and clusters each gray scale pixel point based on the clustering center and each pixel gray scale, thereby achieving the effect of adaptively clustering each gray scale pixel point based on a relatively accurate initial clustering center.

[0063] In order to better understand the specific implementation process of the above clustering, the LBG algorithm is exemplarily described below. Exemplarily, the LBG algorithm is a relatively classic algorithm for image compression based on vector quantization, uses the Lloyd iteration to seek an optimal solution, and can effectively divide a training vector set. The steps of designing vector quantization are as follows: assuming that the training vector set is X={x0, x1,...x M-1},

[0064] 1) Set initial codebook B 0 = {y0 (0) ,y1 (0) ,...y N-1 (0)}, let iteration number n = 0, average distortion D 0 → ∞, and relative error threshold ε (0 ≤ ε ≤ 1);

[0065] 2) According to the nearest neighbor condition, each code word in codebook B n is taken as a centroid (i.e. clustering center), and the training vector set X is divided into N cells S n = {S0 n ,S1 n ,...S N-1 n}, S i n satisfies

[0066] 3) Calculate the average distortion (i.e. gray scale distortion) generated after cell division, the definition of average distortion of each cell is shown in the following formula, wherein the calculation process of min is the process of cell division, y j n is the centroid before updating of the jth cell, D n is the average value of gray scale distance between all x i and y j n of the cell after division;

[0067]

[0068] 4) Determine whether the relative error of the average distortion of the present iteration and the average distortion of the last iteration is less than ε;

[0069]

[0070] 5) If yes, the whole iteration algorithm stops, otherwise, according to the centroid condition, the values of all vectors in each cell in corresponding dimensions are added, and then divided by the number of all vectors in the cell, as the centroid of the cell, the codebook is updated by using the newly determined centroid (i.e. code word) of each cell, let iteration number n = n + 1, and then jump to step 2).

[0071] On this basis, in combination with the application scenarios that the embodiments of the present application can involve, each gray scale pixel point can be taken as a vector to perform training based on the LBG algorithm. Since the LBG algorithm is relatively dependent on an initial codebook, the pixel gray scales of the gray scale pixel points can be sorted here, and a gray scale category split point is determined in the gray scale sorting result based on a pre-set number of categories, and then the gray scale category split point can be taken as an initial codebook to perform LBG iteration.

[0072] Embodiment Three

[0073] Figure 3 is a flowchart of a region detection method provided in Embodiment Three of the present application. The present embodiment is optimized based on the above technical solutions. In the present embodiment, optionally, for a current region in each candidate region, a target region is detected from the candidate regions according to a coupling degree between the region boundary of each candidate region and the gradient edge, and specifically can include: obtaining a first similarity in pixel position between each current pixel point in a current boundary of the current region and each edge pixel point in the gradient edge, and a second similarity in pixel position between each current pixel point and each target pixel point in a target boundary of a candidate region that has been detected as the target region; and determining whether the current region is the target region according to the first similarity, or the first similarity and the second similarity, to achieve detection of the target region. Wherein, the explanations of the same or corresponding terms as in the above embodiments are not repeated here.

[0074] Referring to Figure 3 , the method of the present embodiment can specifically include the following steps:

[0075] S310, clustering each gray scale pixel point according to the pixel gray scale of each gray scale pixel point in the medical image, and taking a region where each gray scale pixel point belonging to the same category is located as a candidate region.

[0076] S320, generating a gradient image corresponding to the medical image according to each pixel gray scale, and determining a gradient edge within the gradient image according to the pixel gradient of each gradient pixel point in the gradient image.

[0077] S330, for a current region in each candidate region, obtaining a first similarity in pixel position between each current pixel point in a current boundary of the current region and each edge pixel point in the gradient edge, and a second similarity in pixel position between each current pixel point and each target pixel point in a target boundary of a candidate region that has been detected as the target region.

[0078] In this context, the current region is the candidate region to be detected at the current moment, the current boundary is the boundary of the current region, and the current pixel is the pixel on the current boundary. The first similarity represents the similarity between each current pixel and each edge pixel at their pixel positions, and it can be determined by the ratio between a first number of current pixels and a second number of current pixels that are the same as or close to a certain edge pixel at their pixel positions. Correspondingly, since the current region can be detected as the target region after each selection from the candidate regions, this means that after the current region is updated, there may be candidate regions that have already been detected as target regions; these candidate regions are the previous current regions. The target boundary is the boundary of the candidate regions that have been detected as target regions, and the target pixel is the pixel on the target boundary. Similarly, the second similarity represents the similarity between each target pixel and each edge pixel at their pixel positions.

[0079] S340. Based on the first similarity or the first similarity and the second similarity, determine whether the current region is the target region in order to achieve the detection of the target region.

[0080] If the first similarity is high, meaning the coupling between each current pixel and each edge pixel is high, then the current region can be identified as the target region. For example, such as... Figure 4a As shown, if the first similarity is determined by the ratio between the first number of current pixels and the second number of current pixels that are the same as or close to an edge pixel in pixel position, then the first similarity can be 50%. Of course, if the first similarity is not very high, the current region cannot be directly considered not to be the target region, because the current boundary of the current region may partially coincide with the gradient edge and partially coincide with the target boundary. In this case, the current region can also be considered the target region. For example, as shown... Figure 4b As shown, if the representation of the second similarity is the same as that of the first similarity, then both the first and second similarities are 25%. That is, it can be determined whether the current region is the target region based on the first similarity, or the first and second similarities. It should be noted that each current region can be detected using the above steps, thus achieving the effect of detecting the target region from among the candidate regions.

[0081] In practical applications, optionally, on the basis of any of the above technical solutions, the detection process of the target region can be determining a region detection order of each candidate region according to a region attribute of the target region to be detected, and screening a current region from each candidate region according to the region detection order, for example, according to the region attribute to determine that the target region is a high-attenuation region, the candidate region corresponding to the high-attenuation region is usually a low-gray candidate region, so the detection can be started from the candidate region with lower region gray; determining whether to take the current region as the target region according to the coupling degree between the region boundary of the current region and the gradient edge, and realizing the detection of the target region according to the determination result; taking the next region of the current region in the region detection order as the current region, and repeatedly executing the step of determining whether to take the current region as the target region according to the coupling degree between the region boundary of the current region and the gradient edge, until the detected candidate region and / or the gradient edge meets a pre-set determination end condition, which can include that the number of the detected candidate region is greater than a pre-set number threshold, the gradient edge no longer exists, and the like, which is not limited here.

[0082] The technical solution of the embodiment of the application determines whether the current region is the target region through the first similarity of each current pixel point in the current boundary and each edge pixel point in the gradient edge in the pixel position, and the second similarity of each current pixel point and each target pixel point in the target boundary of the candidate region detected as the target region, thereby realizing the effect of accurate detection of the target region in different situations.

[0083] On this basis, in order to better understand the specific implementation process of the above target region detection, the following will be described in conjunction with a specific example. Illustratively, taking the target region as a high attenuation region, each category of region gray level from low to high in the LBG classification is traversed step by step, the candidate region of the lowest category that has not been detected is taken as the current region, and the current boundary of the current region is compared with the gradient edge. If the number of current pixel points in the current boundary that are the same as or similar to the edge pixel points on the gradient edge at the pixel position reaches a certain proportion of the total number of current pixel points, it is considered that the current region belongs to a part of the high attenuation region. If the above condition is not met, it can be further judged whether a part of the current boundary is close to the gradient edge and the other part is close to the region boundary (i.e. the target boundary) of the region that has been detected as a high attenuation region. If so, the current region can also be judged as a high attenuation region. Further, the edge pixel points on the gradient edge that are close to the high attenuation region that has been judged are assigned as 0, that is, the edge pixel points occupied by the target boundary are set to 0. The above steps are repeated until the category of the relatively large region gray level in the LBG classification and / or the gradient edge no longer exists in the gradient image are traversed. The reason for this setting is that the former should have completed the detection of the target region in theory, but in actual application, it has not been completed, which may be an error in the region detection process and should be stopped in time. The latter means that all high attenuation regions have been detected, and detection can be stopped at this time.

[0084] Correspondingly, the detection process of the direct exposure region and the high attenuation region can change the region detection order from low to high to high to low, change the lowest category to the highest category, and change the traversal to the category of the relatively large region gray level in the LBG classification to the traversal to the category of the relatively small region gray level in the LBG classification, and the remaining steps are the same, which will not be described here.

[0085] It should be noted that in actual application, it is not necessary to determine in advance whether there is a high attenuation region or a direct exposure region in the medical image. Only when the detection target is a high attenuation region, the detection is performed based on the detection steps related to the high attenuation region, and when the detection target is a direct exposure region, the detection is performed based on the detection steps related to the direct exposure region. That is, the above region detection method can extract the high attenuation region, the direct exposure region, or the high attenuation region and the direct exposure region from the medical image.

[0086] Embodiment Four

[0087] Figure 5A structural block diagram of a region detection device is provided for Embodiment Four of the present application, which is used to execute the region detection method provided by any of the above embodiments. The device and the region detection method of each of the above embodiments belong to the same inventive concept, and the details not described in the embodiment of the region detection device can be referred to the embodiments of the region detection method. Referring to Figure 5 The device can specifically include a candidate region determination module 410, a gradient edge determination module 420, and a target region detection module 430.

[0088] The candidate region determination module 410 is configured to cluster each gray pixel point in the medical image according to the pixel gray of each gray pixel point, and take a region in which each gray pixel point belonging to the same category is located as a candidate region.

[0089] The gradient edge determination module 420 is configured to generate a gradient image corresponding to the medical image according to each pixel gray, and determine a gradient edge in the gradient image according to a pixel gradient of each gradient pixel point in the gradient image.

[0090] The target region detection module 430 is configured to detect a target region from each candidate region according to a coupling degree between a region boundary of each candidate region and the gradient edge.

[0091] Optionally, the candidate region determination module 410 can specifically include:

[0092] A gray category split point determination unit is configured to sort the pixel gray of each gray pixel point in the medical image, and determine a gray category split point in the gray sorting result based on a pre-set category number.

[0093] A gray pixel point clustering unit is configured to take the gray category split point as an initial clustering center, and cluster each gray pixel point based on the clustering center and the pixel gray of each gray pixel point.

[0094] On this basis, the gray pixel point clustering unit can be used to:

[0095] For each gray pixel point, determine a gray distance between the pixel gray of the gray pixel point and each clustering center, and cluster the gray pixel point into a category in which a clustering center corresponding to a smallest gray distance is located.

[0096] Determine a gray distortion according to the smallest gray distance corresponding to each gray pixel point, and determine whether the gray distortion satisfies a pre-set clustering end condition.

[0097] If not, for each category, re-determine a clustering center of the category according to the pixel gray of each gray pixel point belonging to the category after clustering.

[0098] The step of determining the pixel gray scale of the gray scale pixel point and the gray scale distance between each cluster center is repeatedly performed until the gray scale distortion satisfies the cluster end condition, and the clustering ends.

[0099] Optionally, the gradient edge determination module 420 can specifically include:

[0100] The gray scale category region screening unit is configured to screen a gray scale category region corresponding to the region attribute of the target region to be detected from each candidate region according to the region gray scale of each candidate region.

[0101] The first gradient edge determination unit is configured to determine the gradient edge in the gradient image according to the pixel gradient of the gradient pixel point in the gradient image corresponding to each region pixel point in the gray scale category region.

[0102] Optionally, the gradient edge determination module 420 can specifically include:

[0103] The binary image generation unit is configured to sort each gradient pixel point in the gradient image according to the pixel gradient of each gradient pixel point, and screen each gradient pixel point according to the sorting position of each gradient pixel point in the sorting result, to generate a binary image corresponding to the gradient image.

[0104] The second gradient edge determination unit is configured to take the binary edge formed by each binary pixel point in the binary image as the gradient edge of the gradient image.

[0105] Optionally, for the current region in each candidate region, the target region detection module 430 can include:

[0106] The similarity acquisition unit is configured to acquire the first similarity of each current pixel point in the current boundary of the current region and each edge pixel point in the gradient edge in the pixel position, and the second similarity of each current pixel point and each target pixel point in the target boundary of the candidate region detected as the target region in the pixel position.

[0107] The first target region detection unit is configured to determine whether the current region is the target region according to the first similarity, or the first similarity and the second similarity, to realize the detection of the target region.

[0108] Optionally, the target region detection module 430 can specifically include:

[0109] The current region screening unit is configured to determine the region detection order of each candidate region according to the region attribute of the target region to be detected, and screen the current region from each candidate region according to the region detection order.

[0110] The second target region detection unit is configured to determine whether to take the current region as a target region according to the coupling degree between the region boundary of the current region and the gradient edge, and to realize detection of the target region according to the determination result.

[0111] The iteration execution unit is configured to take a next region of the current region in the region detection sequence as the current region, and repeatedly execute the step of determining whether to take the current region as a target region according to the coupling degree between the region boundary of the current region and the gradient edge until the detected candidate region and / or the gradient edge meets a pre-set determination end condition.

[0112] The region detection device provided in the fourth embodiment of the present application clusters each gray pixel point in the medical image according to the pixel gray of each gray pixel point through the candidate region determination module, takes the region in which each gray pixel point belonging to the same category as a candidate region, generates a gradient image corresponding to the medical image according to each pixel gray through the gradient edge determination module, and determines the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image. The target region detection module detects the target region from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge. The above device solves the problem of low detection accuracy of the target region caused by the gray difference between the target region and the human tissue, image noise, and gray transition in the target region, and realizes the effect of accurate detection of the target region by comparing the complete region boundary of the candidate region obtained in this way with the gradient edge of the region boundary which is more likely to be the target region.

[0113] The region detection device provided in the embodiments of the present application can execute the region detection method provided in any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0114] It should be noted that in the embodiments of the above region detection device, each unit and module included is only divided according to the function logic, but is not limited to the above division, as long as the corresponding function can be realized; in addition, the specific name of each functional unit is only for the convenience of mutual differentiation, and does not limit the protection scope of the present application.

[0115] Embodiment five

[0116] Figure 6 The structural schematic diagram of a region detection device provided in the fifth embodiment of the present application is shown in Figure 6 The device includes a memory 510, a processor 520, an input device 530, and an output device 540. The number of processors 520 in the device can be one or more, Figure 6The processor 520 is taken as an example; the memory 510, the processor 520, the input device 530 and the output device 540 in the device can be connected through a bus or other means, Figure 6 The connection through the bus 550 is taken as an example.

[0117] The memory 510 as a kind of computer readable storage medium can be used to store software program, computer executable program and module, such as the program instruction / module (for example, candidate region determination module 410, gradient edge determination module 420 and target region detection module 430 in the region detection device) corresponding to the region detection method in the embodiment of the application. The processor 520 executes the software program, instruction and module stored in the memory 510, so as to execute the various function applications and data processing of the device, that is, to realize the above-mentioned region detection method.

[0118] The memory 510 can mainly include storage program area and storage data area, wherein the storage program area can store operating system, application program required by at least one function; the storage data area can store data created according to the use of the device and the like. In addition, the memory 510 can include high-speed random access memory, and can also include nonvolatile memory, for example, at least one magnetic disk storage device, flash memory device or other nonvolatile solid-state storage device. In some examples, the memory 510 can further include memory arranged remotely relative to the processor 520, which can be connected to the device through a network. Examples of the above-mentioned network include but are not limited to the Internet, intranet, local area network, mobile communication network and combination thereof.

[0119] The input device 530 can be used to receive input digital or character information, and generate key signal input related to user settings and function control of the device. The output device 540 can include display device such as display screen.

[0120] Embodiment six

[0121] The embodiment six of the application provides a storage medium containing computer executable instructions, which are used to execute a region detection method when executed by a computer processor, and the method comprises the following steps:

[0122] According to the pixel gray scale of each gray scale pixel point in the medical image, the gray scale pixel points are clustered, and the region where each gray scale pixel point belonging to the same category is located is taken as a candidate region;

[0123] According to each pixel gray scale, a gradient image corresponding to the medical image is generated, and the gradient edge in the gradient image is determined according to the pixel gradient of each gradient pixel point in the gradient image;

[0124] According to the coupling degree between the region boundary and the gradient edge of each candidate region, a target region is detected from the candidate regions.

[0125] Of course, the storage medium provided by the embodiment of the present application contains computer executable instructions, which are not limited to the method operations as described above, but also can perform the related operations in the region detection method provided by any embodiment of the present application.

[0126] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary universal hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. According to such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk or an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in various embodiments of the present application.

[0127] Note that the above are only the preferred embodiments of the present application and the technical principles applied. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and those skilled in the art can make various obvious changes, readjustments and substitutions without departing from the scope of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A region detection method characterized by, The method comprises the steps of: clustering each gray pixel point in a medical image according to the pixel gray of each gray pixel point, and taking the region where each gray pixel point belonging to the same category is located as a candidate region; generating a gradient image corresponding to the medical image according to each pixel gray, and determining a gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image; detecting a target region from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge; for a current region in each candidate region, the step of detecting a target region from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge comprises: obtaining a first similarity of each current pixel point in the current boundary of the current region and each edge pixel point in the gradient edge in the pixel position, and a second similarity of each current pixel point and each target pixel point in the target boundary of the candidate region detected as the target region in the pixel position; wherein the current region is a candidate region detected at the current time; judging whether the current region is the target region according to the first similarity and the second similarity to realize the detection of the target region.

2. The method of claim 1, wherein, The step of clustering each gray pixel point in a medical image according to the pixel gray of each gray pixel point comprises: sorting the pixel gray of each gray pixel point in the medical image, and determining a gray category split point in the gray sorting result based on a pre-set number of categories; taking the gray category split point as an initial clustering center, and clustering each gray pixel point based on the clustering center and the pixel gray of each gray pixel point.

3. The method of claim 2, wherein, The step of clustering each gray pixel point based on the clustering center and the pixel gray of each gray pixel point comprises: for each gray pixel point, determining the gray distance between the pixel gray of the gray pixel point and each clustering center, and clustering the gray pixel point into the category where the clustering center corresponding to the smallest gray distance is located; determining a gray distortion according to the smallest gray distance corresponding to each gray pixel point, and judging whether the gray distortion meets a pre-set clustering end condition; if not, for each category, re-determining the clustering center of the category according to the pixel gray of each gray pixel point in the category after clustering; repeating the step of determining the gray distance between the pixel gray of the gray pixel point and each clustering center until the gray distortion meets the clustering end condition, and ending the clustering.

4. The method of claim 1, wherein, The step of determining a gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image comprises: screening a gray category region corresponding to the region attribute of the target region to be detected from each candidate region according to the region gray of each candidate region; Determine the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image.

5. The method of claim 1, wherein, The determination of the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image comprises: Sort each gradient pixel point according to the pixel gradient of each gradient pixel point in the gradient image, and screen each gradient pixel point according to the sorting position of each gradient pixel point in the sorting result to generate a binary image corresponding to the gradient image; The binary edge formed by each binary pixel point in the binary image is taken as the gradient edge of the gradient image.

6. The method of claim 1, wherein, The target region is detected from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge, comprising: Determine the region detection order of each candidate region according to the region attribute of the target region to be detected, and screen the current region from each candidate region according to the region detection order; Determine whether the current region is taken as the target region according to the coupling degree between the region boundary of the current region and the gradient edge, and realize the detection of the target region according to the judgment result; The next region of the current region in the region detection order is taken as the current region, and the step of determining whether the current region is taken as the target region according to the coupling degree between the region boundary of the current region and the gradient edge is repeatedly executed until the candidate region and / or the gradient edge detected satisfies the pre-set judgment end condition.

7. An area detection device, characterized by Comprise: The candidate region determination module is used for clustering each gray pixel point in the medical image according to the pixel gray scale of each gray pixel point, and taking the region where each gray pixel point belonging to the same category is located as a candidate region; The gradient edge determination module is used for generating a gradient image corresponding to the medical image according to each pixel gray scale, and determining the gradient edge in the gradient image according to the pixel gradient of each gradient pixel point in the gradient image; The target region detection module is used for detecting a target region from each candidate region according to the coupling degree between the region boundary of each candidate region and the gradient edge. For the current region in each candidate region, the target region detection module comprises: The similarity obtaining unit is used for obtaining the first similarity of each current pixel point in the current boundary of the current region and each edge pixel point in the gradient edge in the pixel position, and the second similarity of each current pixel point and each target pixel point in the target boundary of the candidate region detected as the target region in the pixel position; wherein the current region is the candidate region detected at the current time; The first target region detection unit is used for judging whether the current region is the target region according to the first similarity, or the first similarity and the second similarity, to realize the detection of the target region.

8. An area detection device, characterized by Comprise: One or more processors; Memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a region detection method as claimed in any one of claims 1-6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, which is executed by a processor, implements a region detection method as claimed in any one of claims 1-6.

Images