Method for extracting morphological features of the base of the tongue area for patients with cervical HPV infection

Abstract

The present application relates to the technical field of image analysis, in particular to a tongue root area morphological feature extraction method for cervical HPV infected patients, which comprises: acquiring each frame of target tongue image of the target patient in the process of stretching and retracting tongue action, and screening out tongue retraction image and tongue extension image based on the morphology of tongue area in the image; constructing standard retracted tongue root area and standard extended tongue root area; performing physiological wrinkle feature elimination to determine texture stability factor; performing saliva highlight reflection feature elimination to determine synergistic feature factor; constructing hyperplastic tissue feature factor according to the texture stability factor and the synergistic feature factor. The present application analyzes the target tongue image, to a certain extent, eliminates normal features such as physiological wrinkle and saliva highlight reflection, realizes the extraction of the abnormal morphological feature of hyperplastic tissue feature, and improves the rationality of the extraction of the hyperplastic tissue feature, so that the subsequent positioning of the hyperplastic tissue area can be facilitated.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, specifically to a method for extracting morphological features of the tongue base region in patients with cervical HPV infection. Background Technology

[0002] Morphological features of the tongue root region often include: physiological folds, salivary highlighting, and proliferative tissue features. Physiological folds and salivary highlighting are usually normal features, while proliferative tissue features are usually abnormal. Generally, abnormal features are more valuable for analysis; therefore, extracting proliferative tissue features is often crucial. Currently, the common method for feature extraction is to distinguish specific regions based on their grayscale values, thereby achieving feature extraction for those specific regions.

[0003] However, when identifying proliferative tissue based on different grayscale values ​​to extract its features, the following technical problems often arise: In reality, the grayscale difference between hyperplastic tissue regions and normal tissue regions is often not significant. Therefore, when identifying hyperplastic tissue, if only the difference in grayscale values ​​is considered, it may lead to misjudgment of hyperplastic tissue pixels, resulting in poor accuracy of hyperplastic tissue identification and consequently poor rationality of hyperplastic tissue feature extraction. Summary of the Invention

[0004] To address the technical problem of poor rationality in the extraction of features from proliferating tissue, this invention proposes a method for extracting morphological features of the tongue base region in patients with cervical HPV infection.

[0005] In a first aspect, the present invention provides a method for extracting morphological features of the tongue base region in patients with cervical HPV infection, the method comprising: Acquire each frame of the target tongue body image during the target patient's tongue extension and retraction movements, and based on the morphology of the tongue body region in the target tongue body image, filter out the tongue retraction image and the tongue extension image from all the target tongue body images; Based on images of tongue retraction and tongue extension, standard retracted tongue root regions and standard extended tongue root regions are constructed. Pixels that match each pixel in the standard extended tongue root region are selected from the standard retracted tongue root region, thus obtaining the matching pixels corresponding to each pixel in the standard retracted tongue root region. The standard retracted tongue root region is divided into blocks to obtain initial blocks, and the target block is selected from all the initial blocks; Physiological wrinkle features are removed based on the similarity between the gradient values ​​of pixels within each target block and their corresponding matching pixels, and the texture stabilization factor for each target block is determined. Saliva specular reflection features are removed based on the texture stabilization factors corresponding to all target blocks, and the co-feature factors corresponding to each target block are determined. Based on the texture stability factor and synergistic feature factor corresponding to all target blocks, construct the proliferative tissue feature factor.

[0006] In conjunction with the first aspect above, in one possible implementation, the step of filtering out tongue retraction images and tongue extension images from all target tongue images based on the morphology of the tongue region in the target tongue image includes: The tongue region is selected from each frame of the target tongue image; The aspect ratio of the minimum bounding rectangle of the tongue region in each frame of the target tongue image is determined as the morphological representative feature factor corresponding to each frame of the target tongue image. The target tongue image with the largest corresponding morphological characteristic factor is selected from all target tongue images and used as the tongue extension image; The target tongue image with the smallest corresponding morphological characteristic factor is selected from all target tongue images and used as the tongue retraction image.

[0007] In conjunction with the first aspect above, in one possible implementation, constructing a standard retracted tongue root region and a standard extended tongue root region based on the tongue retraction image and the tongue extension image includes: Based on the geometric center of the tongue region in the tongue retraction image, the initial retracted tongue root region is extracted from the tongue retraction image; Based on the geometric center of the tongue region in the tongue extension image, the initial extended tongue root region is extracted from the tongue extension image; Using a bilinear interpolation algorithm, the initial retracted tongue root region and the initial extended tongue root region are scaled to the same size to obtain the standard retracted tongue root region and the standard extended tongue root region.

[0008] In conjunction with the first aspect above, in one possible implementation, the step of extracting an initial retracted tongue root region from the tongue retraction image based on the geometric center of the tongue region in the tongue retraction image includes: The main axis of the tongue is obtained by least squares fitting of the contour of the tongue region in the tongue retraction image, and is denoted as the main axis of tongue retraction. The geometric center of the tongue region in the tongue retraction image is translated by a preset offset along the main axis of tongue retraction toward the root of the tongue, and the position point obtained after translation is determined as the target position point; Centered on the target location point, construct a square region with a side length equal to the preset length, and denote it as the initial retraction tongue root region.

[0009] In conjunction with the first aspect above, in one possible implementation, the step of selecting pixels from the standard extended tongue root region that match each pixel in the standard retracted tongue root region to obtain the matching pixel corresponding to each pixel in the standard retracted tongue root region includes: Pixels with gray values ​​greater than a preset gray threshold are selected from the standard retracted tongue root area and used as retracted highlight points. Pixels with gray values ​​less than or equal to a preset gray threshold are selected from the standard retracted tongue root area and used as non-retracted highlight points. Using the LK optical flow algorithm, pixels that match each non-retracted highlight point are selected from the standard extended tongue root area and used as the matching pixels corresponding to each non-retracted highlight point. Based on the matching pixels corresponding to non-retracted highlight points within the preset neighborhood of each retracted highlight point, the matching pixels corresponding to each retracted highlight point are selected.

[0010] In conjunction with the first aspect above, in one possible implementation, the step of filtering out the matching pixels corresponding to each retracted highlight point based on the matching pixels corresponding to non-retracted highlight points within a preset neighborhood of each retracted highlight point includes: Any retracted highlight point is designated as a marked highlight point, and a displacement vector corresponding to each non-retracted highlight point is constructed based on each non-retracted highlight point and its corresponding matching pixel point. The mean value of the displacement vectors corresponding to all non-retracted highlight points in the preset neighborhood of the marked highlight point is determined as the target displacement vector. From the standard extended tongue root region, select the position points whose displacement vector between them and the marked highlight points is equal to the target displacement vector, and use them as reference position points; The pixel closest to the reference position point is determined as the matching pixel corresponding to the marked highlight point.

[0011] In conjunction with the first aspect above, in one possible implementation, the step of removing physiological wrinkle features based on the similarity between the gradient values ​​of pixels within each target block and their corresponding matching pixels, and determining the texture stabilization factor for each target block, includes: The gradient values ​​corresponding to all pixels within each target block are used to form the original gradient value sequence for each target block. The gradient values ​​of the matching pixels corresponding to all pixels in each target block are used to form the matching gradient value sequence for each target block. The texture stabilization factor for each target block is determined based on the Pearson correlation coefficient between the original gradient value sequence and the matching gradient value sequence corresponding to each target block.

[0012] In conjunction with the first aspect above, in one possible implementation, the step of removing saliva specular reflection features based on the texture stabilization factors corresponding to all target blocks, and determining the collaborative feature factor corresponding to each target block, includes: Based on the texture stability factor corresponding to the target block, determine the target association value between target blocks; Based on the target association values ​​between target blocks, an organizational spatial association matrix is ​​constructed with the target blocks as nodes, where the target association values ​​are the elements in the organizational spatial association matrix. Based on the organizational spatial correlation matrix, a target degree matrix is ​​constructed, where the cumulative value of each row of elements in the organizational spatial correlation matrix is ​​used as the element on the diagonal of the target degree matrix. Construct a target identity matrix, wherein the target identity matrix, the organization space correlation matrix, and the target degree matrix have the same size; Generate a Laplace matrix based on the target identity matrix, the organization space correlation matrix, and the target degree matrix; Perform eigenvalue decomposition on the Laplacian matrix to obtain multiple eigenvalues ​​and their corresponding eigenvectors; Normalize the eigenvector corresponding to the second smallest eigenvalue obtained by matrix eigenvalue decomposition to obtain the co-eigenvector; The element corresponding to each target block in the collaborative feature vector is determined as the collaborative feature factor for each target block.

[0013] In conjunction with the first aspect above, in one possible implementation, determining the target association value between target blocks based on the texture stability factor corresponding to the target block includes: Any target block is identified as a marked region. From all target blocks, target blocks whose distance from the marked region is not 0 and whose distance from the marked region is less than or equal to a preset distance threshold are selected and identified as candidate blocks. Each target block other than the candidate blocks in all target blocks is identified as a temporary block, and the target association value between the marked area and each temporary block is set to a preset association value; Based on the difference between the texture stability factor corresponding to the marked region and the texture stability factor corresponding to each candidate block, and the distance between the marked region and each candidate block, the target correlation value between the marked region and each candidate block is determined.

[0014] In conjunction with the first aspect above, in one possible implementation, the step of selecting the target block from all initial blocks includes: Select the initial blocks whose center point is located within the tongue area from all initial blocks, and use them as the target blocks.

[0015] Secondly, this invention provides a system for extracting morphological features of the tongue base region in patients with cervical HPV infection, the system comprising: The image acquisition and filtering module is used to acquire each frame of the target tongue image during the tongue extension and retraction movement of the target patient, and to filter out the tongue retraction image and the tongue extension image from all the target tongue images based on the morphology of the tongue region in the target tongue image. The standard tongue root region construction module is used to construct the standard retracted tongue root region and the standard extended tongue root region based on the tongue retraction image and the tongue extension image; The pixel matching module is used to filter out pixels that match each pixel in the standard extended tongue root area from the standard retracted tongue root area, and obtain the matching pixels corresponding to each pixel in the standard retracted tongue root area. The block division and filtering module is used to divide the standard retracted tongue root region into blocks to obtain initial blocks, and to filter out target blocks from all initial blocks. The texture stabilization factor determination module is used to remove physiological wrinkle features based on the similarity between the gradient values ​​of pixels in each target block and their corresponding matching pixels, and to determine the texture stabilization factor for each target block. The collaborative feature factor determination module is used to remove saliva specular reflection features based on the texture stabilization factors corresponding to all target blocks and determine the collaborative feature factor corresponding to each target block. The proliferative tissue feature extraction module is used to construct proliferative tissue feature factors based on the texture stabilization factor and collaborative feature factor corresponding to all target blocks.

[0016] The present invention has the following beneficial effects: This invention provides a method for extracting morphological features of the tongue base region in patients with cervical HPV infection. By analyzing the target tongue image, it eliminates normal features such as physiological folds and saliva highlights to a certain extent, enabling the extraction of abnormal morphological features, namely hyperplastic tissue characteristics. This solves the technical problem of poor rationality in hyperplastic tissue feature extraction, improving its rationality and facilitating subsequent localization of hyperplastic tissue areas. In reality, physiological folds are often physical protrusions formed by the overlap of normal mucosa when the tongue muscles contract inward and are compressed. These protrusions tend to flatten and disappear when the tongue is stretched outward. Therefore, by analyzing images of tongue retraction (representing tongue contraction) and tongue extension (representing tongue stretching outward), the characteristic patterns caused by these protrusions can be identified and eliminated to a certain extent. Furthermore, saliva highlights can be removed, allowing for relatively objective extraction of hyperplastic tissue characteristic factors, thus improving the rationality of hyperplastic tissue feature extraction and aiding in the localization of hyperplastic tissue. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to the present invention. Figure 2 This is a schematic diagram of the composition structure of a tongue base region morphological feature extraction system for patients with cervical HPV infection according to the present invention. Figure 3 This is a schematic diagram of the structure of a computer device according to the present invention. Detailed Implementation

[0018] refer to Figure 1 This paper illustrates the flowchart of a method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to the present invention. The method includes the following steps: Step S1: Acquire each frame of the target tongue image during the tongue extension and retraction movement of the target patient, and based on the morphology of the tongue region in the target tongue image, filter out the tongue retraction image and the tongue extension image from all the target tongue images.

[0019] The target patient can be a cervical HPV (Human Papillomavirus) infected patient for whom morphological features of the abnormal tongue base region (specifically, tongue base hyperplasia tissue characteristics) are to be extracted. The tongue extension / retraction movement can be the process of stretching the tongue outwards and then retracting it. The target tongue image can be a grayscale image of the upper surface of the target patient's tongue. The tongue region can represent the upper surface of the tongue. The tongue retraction image can represent the relaxed state of the tongue retracted. The tongue extension image can represent the taut state of the tongue stretched outwards.

[0020] As an example, this step may include the following steps: The first step is to acquire each frame of the target tongue image during the target patient's tongue extension and retraction movements.

[0021] For example, a camera can be used to capture each frame of grayscale image of the upper surface of the target patient's tongue during the tongue extension and retraction movements, which can then be used as the target tongue image.

[0022] The second step is to select the tongue region from each frame of the target tongue image.

[0023] For example, existing technologies such as neural network technology or edge detection algorithms can be used to identify the tongue region in a target tongue image.

[0024] The third step is to determine the aspect ratio of the minimum bounding rectangle of the tongue region in each frame of the target tongue image as the morphological representative feature factor corresponding to each frame of the target tongue image.

[0025] For example, the formula for determining the morphological representative feature factor corresponding to the target tongue image can be: ; in, is the morphological characteristic factor corresponding to the j-th frame of the target tongue image. j is the frame number of the target tongue image. It is the length of the minimum bounding rectangle of the tongue region in the j-th frame of the target tongue image. It is the width of the minimum bounding rectangle of the tongue region in the j-th frame of the target tongue image.

[0026] It should be noted that, It can characterize the degree of outward extension of the tongue muscles represented by the j-th frame of the target tongue image. The larger the value, the greater the degree of outward extension of the tongue represented by the j-th frame of the target tongue image.

[0027] The fourth step is to select the target tongue image with the largest corresponding morphological characteristic factor from all target tongue images and use it as the tongue extension image.

[0028] The fifth step is to select the target tongue image with the smallest corresponding morphological characteristic factor from all target tongue images, and use it as the tongue retraction image.

[0029] Step S2: Based on the images of tongue retraction and tongue extension, construct the standard retracted tongue root region and the standard extended tongue root region.

[0030] The standard retracted tongue root region and the standard extended tongue root region can be the same size. The standard retracted tongue root region represents the area in a retracted tongue image where the tongue root is relatively centrally located and covers the tongue body. The standard extended tongue root region represents the area in a stretched tongue image where the tongue root is relatively centrally located and covers the tongue body.

[0031] As an example, this step may include the following steps: The first step, extracting the initial retracted tongue root region from the tongue retraction image based on the geometric center of the tongue region in the tongue retraction image, may include the following sub-steps: The first sub-step involves using least squares fitting to obtain the main axis of the tongue body in the tongue retraction image, denoted as the main axis of tongue retraction.

[0032] Among them, the main axis of the tongue can represent the center line of the tongue region.

[0033] The second sub-step involves shifting the geometric center of the tongue region in the tongue retraction image along the main axis of tongue retraction toward the root of the tongue by a preset offset, and determining the position point obtained after the shift as the target position point.

[0034] The preset offset can be a pre-set offset to be translated, such as 0.25 times the length of the longest side of the smallest bounding rectangle of the tongue region in the tongue retraction image. In reality, the geometric center of the tongue region is often located at the middle and front of the tongue. This embodiment of the invention mainly analyzes the situation of the tongue root. Therefore, to facilitate the analysis of the tongue root, the geometric center of the tongue region can be translated to bring the tongue root region as close as possible to the central position to be analyzed. The target position point can represent a new center point close to the tongue root region.

[0035] The third sub-step involves constructing a square region with a side length equal to the preset length, centered on the target location point, and denoted as the initial retraction tongue root region.

[0036] The initial retracted tongue root region can be the area in the tongue retraction image where the tongue root is relatively centered and covers the tongue body. The preset length can be a pre-set length, such as the sum of the length of the longest side of the smallest bounding rectangle of the tongue body region in the tongue retraction image and the target offset. The target offset can be equal to twice the preset offset.

[0037] It should be noted that the initial retraction of the tongue root region often needs to cover the tongue body region in the tongue body retraction image. Therefore, the preset length can be greater than the length of the long side of the smallest bounding rectangle of the tongue body region in the tongue body retraction image.

[0038] The second step is to extract the initial extended tongue root region from the tongue extension image based on the geometric center of the tongue region in the tongue extension image.

[0039] The initial extended tongue root region can be the area in the tongue extension image where the tongue root is located in the center and covers the tongue body.

[0040] It should be noted that the method for obtaining the initial extended tongue root region can refer to the method for obtaining the initial retracted tongue root region, and will not be repeated here.

[0041] The third step involves using bilinear interpolation to scale the initial retracted tongue root region and the initial extended tongue root region to the same size, resulting in the standard retracted tongue root region and the standard extended tongue root region.

[0042] For example, a bilinear interpolation algorithm can be used to uniformly scale the initial retracted tongue root region and the initial extended tongue root region to a fixed size of 100×100 pixels, and the scaled initial retracted tongue root region is recorded as the standard retracted tongue root region, and the scaled initial extended tongue root region is recorded as the standard extended tongue root region.

[0043] Step S3: Select pixels from the standard extended tongue root area that match each pixel in the standard retracted tongue root area to obtain the matching pixels corresponding to each pixel in the standard retracted tongue root area.

[0044] As an example, this step may include the following steps: The first step is to select pixels with gray values ​​greater than a preset gray value threshold from the standard retracted tongue root area and use them as retracted highlight points.

[0045] The preset grayscale threshold can be a pre-set grayscale boundary used to determine the highlight point, such as 240. The retracted highlight point can represent the highlight pixels within the standard retracted tongue root region.

[0046] The second step is to select pixels with gray values ​​less than or equal to a preset gray value threshold from the standard retracted tongue root area, and use them as non-retracted highlight points.

[0047] Among them, non-retracted highlight points can characterize non-highlight pixels within the standard retracted tongue root region.

[0048] The third step involves using the LK (Lucas Kanade) optical flow algorithm to select pixels that match each non-retracted highlight point from the standard extended tongue root region, which are then used as the matching pixels for each non-retracted highlight point.

[0049] It should be noted that the LK optical flow algorithm often relies on the assumption of constant brightness. However, the highlights formed by the reflection of saliva in the mouth often undergo random shifts in brightness and position during movement. Directly inputting an image with highlights into the LK optical flow algorithm may cause local mapping vector divergence or failure. Therefore, this embodiment of the invention introduces a highlight isolation mechanism, which uses the LK optical flow algorithm to match only non-retracted highlights.

[0050] The fourth step is to filter out the matching pixels corresponding to each shrink highlight point based on the matching pixels corresponding to the non-shrink highlight points in the preset neighborhood of each shrink highlight point.

[0051] The preset neighborhood corresponding to the shrunken highlight can be a neighborhood containing at least one non-shrunken highlight. For example, a 3×3 neighborhood can be the smallest neighborhood. If the 3×3 neighborhood corresponding to the shrunken highlight contains at least one non-shrunken highlight, then the preset neighborhood corresponding to the shrunken highlight can be a 3×3 neighborhood. If the 3×3 neighborhood corresponding to the shrunken highlight does not contain a non-shrunken highlight, then the neighborhood is expanded, and it is determined whether the 5×5 neighborhood corresponding to the shrunken highlight contains a non-shrunken highlight. If the 5×5 neighborhood corresponding to the shrunken highlight contains at least one non-shrunken highlight, then the preset neighborhood corresponding to the shrunken highlight can be a 5×5 neighborhood. If the 5×5 neighborhood corresponding to the shrunken highlight does not contain a non-shrunken highlight, then the neighborhood is expanded again until the finally expanded neighborhood contains at least one non-shrunken highlight, and the finally obtained neighborhood is taken as the preset neighborhood corresponding to the shrunken highlight.

[0052] For example, selecting the matching pixels corresponding to each retracted highlight point may include the following sub-steps: The first sub-step involves identifying any retracted highlight as a marked highlight, and constructing a displacement vector for each non-retracted highlight based on each non-retracted highlight and its corresponding matching pixel.

[0053] Among them, the displacement vector corresponding to the non-retracted highlight point can characterize the motion of the non-retracted highlight point.

[0054] For example, the LK optical flow algorithm can be used to obtain the displacement vector between each non-retracted highlight point and its corresponding matching pixel point, and denoted as the displacement vector corresponding to each non-retracted highlight point.

[0055] The second sub-step is to determine the mean value of the displacement vectors corresponding to all non-retracted highlight points within the preset neighborhood of the marked highlight point as the target displacement vector.

[0056] The third sub-step involves selecting, from the standard extended tongue root region, locations whose displacement vectors relative to the marked highlight points are equal to the target displacement vectors, and using these locations as reference locations.

[0057] The fourth sub-step involves identifying the pixel closest to the aforementioned reference position as the matching pixel corresponding to the marked highlight point.

[0058] Step S4: Divide the standard retracted tongue root region into blocks to obtain initial blocks, and select the target block from all the initial blocks.

[0059] As an example, this step may include the following steps: The first step is to divide the standard retracted tongue root region into blocks to obtain the initial blocks.

[0060] The initial block can be a 5×5 square area.

[0061] For example, the standard retracted tongue root region can be divided into equal parts, and the resulting region can be denoted as the initial block.

[0062] The second step is to select the initial blocks whose center point is located within the tongue region from all the initial blocks, and use them as target blocks. Here, the tongue region can be the tongue region in the tongue retraction image.

[0063] Step S5: Based on the similarity between the gradient values ​​of pixels in each target block and their corresponding matching pixels, physiological wrinkle features are removed, and the texture stabilization factor corresponding to each target block is determined.

[0064] It should be noted that normal physiological wrinkles tend to flatten out gradually during muscle stretching, causing a severe attenuation or loss of the image gradient of their surface texture; while real pathological hyperplastic granules are usually denser (harder in texture) and tend to maintain their original shape when stretched by muscles, and the distribution pattern of their image gradient often shows a high degree of consistency before and after stretching.

[0065] As an example, this step may include the following steps: The first step is to construct the original gradient value sequence for each target block by taking the gradient values ​​of all pixels within each target block.

[0066] For example, any target block can be designated as a marked region. All pixels within the marked region are sorted from top to bottom and left to right to obtain a pixel sequence corresponding to the marked region. The gradient values ​​of all pixels in this pixel sequence are then used to construct the original gradient value sequence for the marked region. Each pixel in the marked region's pixel sequence corresponds one-to-one with an element in its original gradient value sequence, and vice versa.

[0067] The second step is to construct a sequence of matching gradient values ​​for each target block by taking the gradient values ​​of the matching pixels corresponding to all pixels within each target block.

[0068] In this case, the order of elements in the matching gradient value sequence and the original gradient value sequence corresponding to the same target block can be the same.

[0069] The third step is to determine the texture stabilization factor for each target block based on the Pearson correlation coefficient between the original gradient value sequence and the matching gradient value sequence corresponding to each target block.

[0070] For example, the formula for determining the texture stabilization factor corresponding to the target block can be: ; in, It is the texture stabilization factor corresponding to the i-th target block. i is the index of the target block. It is a function that takes the maximum value. It is the Pearson correlation coefficient between the original gradient value sequence and the matching gradient value sequence corresponding to the i-th target block.

[0071] It should be noted that when The larger the value, the more likely the gradient distribution of the area represented by the i-th target block is to remain unchanged during muscle stretching. This also indicates that the area represented by the i-th target block is more likely to conform to the characteristic pattern of pathological hyperplastic granules, and can, to some extent, exclude the interference of physiological wrinkles.

[0072] Step S6: Based on the texture stabilization factor corresponding to all target blocks, remove the saliva specular reflection feature and determine the collaborative feature factor corresponding to each target block.

[0073] As an example, this step may include the following steps: The first step, determining the target association values ​​between target blocks based on the texture stabilization factor corresponding to the target blocks, may include the following sub-steps: The first sub-step involves identifying any target block as a marked region, and then selecting target blocks from all target blocks whose distance from the marked region is not 0 and whose distance from the marked region is less than or equal to a preset distance threshold, and identifying these as candidate blocks.

[0074] The distance between two regions can be equal to the distance between their geometric centers, which can be represented by the number of pixels between their geometric centers. The preset distance threshold can be a pre-defined threshold, such as 5.

[0075] It should be noted that the area with a distance of 0 from the marked area is the marked area itself.

[0076] The second sub-step involves identifying each target block other than the candidate blocks as a temporary block and setting the target association value between the marked area and each temporary block to a preset association value.

[0077] The preset associated value can be a pre-set associated value, such as 0.

[0078] It should be noted that self-loops are generally not allowed in the construction of the Laplacian matrix. Therefore, the target association value between the marked regions themselves can be set to 0. For target blocks whose distance from each other is greater than a preset distance threshold, the distance between these target blocks is relatively far and their connectivity is often low. Therefore, the target association value between these target blocks can be set to 0.

[0079] The third sub-step involves determining the target correlation value between the marked region and each candidate block based on the difference between the texture stability factor corresponding to the marked region and the texture stability factor corresponding to each candidate block, as well as the distance between the marked region and each candidate block.

[0080] For example, the formula for determining the target association value between the marked region and the candidate block can be: ; in, This is the target association value between the marked region and the a-th candidate block. 'a' is the sequence number of the candidate block. It is an exponential function with the natural constant as its base. It is the distance between the marked region and the a-th candidate block. It is the side length of the standard retracted tongue root region. It is a normalized value. It is an absolute value function. E is the texture stabilization factor corresponding to the marked region. It is the texture stability factor corresponding to the a-th candidate block.

[0081] It should be noted that when The smaller the value, the closer the spatial distance between the marked region and the a-th candidate block, and the higher its potential connectivity. The smaller the value, the more similar the pathological stiffness of the marked region and the tissue corresponding to the a-th candidate block may be. Therefore, when A larger value usually indicates a higher correlation between the marked region and the a-th candidate block.

[0082] The second step is to construct an organizational spatial association matrix based on the target association values ​​between target blocks, using the target blocks as nodes.

[0083] The target association values ​​can be elements in the organizational spatial association matrix. The organizational spatial association matrix can be an N×N matrix, where N is the number of target blocks. For example, the element in the first row and first column of the organizational spatial association matrix can be the target association value between the first target block and itself; the element in the first row and second column can be the target association value between the first target block and the second target block; and so on, to construct a complete organizational spatial association matrix.

[0084] The third step is to construct the target degree matrix based on the spatial correlation matrix of the organization.

[0085] The target degree matrix can be a diagonal matrix. The sum of the elements in each row of the organizational spatial correlation matrix can be used as the diagonal element of the target degree matrix. For example, the sum of the elements in the first row of the organizational spatial correlation matrix can be used as the first diagonal element of the target degree matrix; the sum of the elements in the second row of the organizational spatial correlation matrix can be used as the second diagonal element of the target degree matrix; and so on, to obtain the complete target degree matrix.

[0086] The fourth step is to construct the target identity matrix.

[0087] The target identity matrix, the organizational spatial correlation matrix, and the target degree matrix can all have the same size.

[0088] For example, an identity matrix with the same size as the organizational spatial correlation matrix can be constructed as the target identity matrix.

[0089] The fifth step is to generate the Laplace matrix based on the target identity matrix, the organization space correlation matrix, and the target degree matrix.

[0090] For example, the formula for determining the Laplace matrix can be: ; Where L is the Laplace matrix, H is the target identity matrix, B is the target degree matrix, and F is the organizational spatial correlation matrix.

[0091] The sixth step involves using the eigenvalue decomposition (EVD) algorithm to decompose the Laplacian matrix into eigenvalues, obtaining multiple eigenvalues ​​and their corresponding eigenvectors.

[0092] The eigenvalues ​​obtained by performing eigenvalue decomposition on the Laplacian matrix can be scalar values ​​reflecting the topological connectivity of the graph network, and are often all non-negative. The second smallest eigenvalue, also known as algebraic connectivity, often represents the ease with which the network can be divided into two parts. The eigenvector corresponding to the second smallest eigenvalue is called the Fiedler vector. The number of elements in this vector can be equal to the number of target blocks, and its elements often correspond one-to-one with the target blocks. Target blocks in connected clusters (patches of lesion-proliferating tissue) will have very similar values ​​in this vector; while isolated nodes (saliva-reflective noise) will have values ​​that deviate from the majority.

[0093] The seventh step is to normalize the eigenvector corresponding to the second smallest eigenvalue obtained by matrix eigenvalue decomposition to obtain the co-eigenvector.

[0094] For example, the formula for obtaining the elements corresponding to the collaborative feature vector can be: ; in, It is the t-th element in the collaborative feature vector. It is an exponential function with the natural constant as its base. It is an absolute value function. is the t-th element in the eigenvector corresponding to the second smallest eigenvalue. k is the median of all elements in the eigenvector corresponding to the second smallest eigenvalue. It is the standard deviation of all elements within the eigenvector corresponding to the second smallest eigenvalue. If A value of 0 usually indicates the absence of isolated nodes, and all elements in the collaborative feature vector can be set to 1. The formula corresponding to the elements in the collaborative feature vector only applies to... Cases where the value is not equal to 0.

[0095] It should be noted that, The closer a value is to 1, the closer the t-th element in the eigenvector corresponding to the second smallest eigenvalue is to the median, which often indicates that it is more likely to correspond to proliferative tissue. When the value is closer to 0, it often indicates that the t-th element in the eigenvector corresponding to the second smallest eigenvalue is further away from the median, which often indicates that it is more likely to correspond to an isolated node (saliva-reflective noise).

[0096] The eighth step is to determine the corresponding element of each target block in the above collaborative feature vector as the collaborative feature factor for each target block.

[0097] Step S7: Construct proliferative tissue feature factors based on the texture stabilization factors and collaborative feature factors corresponding to all target blocks.

[0098] As an example, the formula for determining the characteristic factors of proliferating tissue can be: ; G is a characteristic factor of proliferating tissue. It is the average of the texture stability factors corresponding to all target blocks. It is the mean of the collaborative feature factors corresponding to all target blocks.

[0099] It should be noted that G can characterize the features of proliferating tissue. The larger the value, the more likely the target patient's tongue is to conform to the characteristics of proliferating tissue, and the more likely the target patient's tongue is to have proliferating tissue. To a certain extent, it can eliminate the interference of physiological folds and the high reflectivity of saliva.

[0100] Optionally, the product between the texture stabilization factor and the co-feature factor corresponding to each target block can be determined as the block proliferation feature factor corresponding to each target block. When the block proliferation feature factor corresponding to the target block is larger, it often indicates that the target block is more likely to represent proliferation tissue, which can help to locate proliferation tissue to a certain extent.

[0101] Based on the same inventive concept as the above-described method embodiments, this invention provides a system for extracting morphological features of the tongue base region in patients with cervical HPV infection. This system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps of a method for extracting morphological features of the tongue base region in patients with cervical HPV infection. Specifically, refer to... Figure 2 A system for extracting morphological features of the tongue base region in patients with cervical HPV infection may specifically include: The image acquisition and filtering module 201 is used to acquire each frame of the target tongue image during the tongue extension and retraction movement of the target patient, and to filter out the tongue retraction image and the tongue extension image from all the target tongue images based on the morphology of the tongue region in the target tongue image. The standard tongue root region construction module 202 is used to construct the standard retracted tongue root region and the standard extended tongue root region based on the tongue retraction image and the tongue extension image; The pixel matching module 203 is used to filter out pixels that match each pixel in the standard retracted tongue root area from the standard extended tongue root area, so as to obtain the matching pixel corresponding to each pixel in the standard retracted tongue root area. The block division and filtering module 204 is used to divide the standard retracted tongue root region into blocks to obtain initial blocks, and to filter out target blocks from all initial blocks. The texture stabilization factor determination module 205 is used to remove physiological wrinkle features based on the similarity between the gradient values ​​of pixels in each target block and their corresponding matching pixels, and to determine the texture stabilization factor for each target block. The collaborative feature factor determination module 206 is used to remove saliva highlight reflection features based on the texture stability factors corresponding to all target blocks and determine the collaborative feature factor corresponding to each target block. The proliferative tissue feature extraction module 207 is used to construct proliferative tissue feature factors based on the texture stabilization factor and collaborative feature factor corresponding to all target blocks.

[0102] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. For example, as shown... Figure 3 As shown, the computer device 300 includes: a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302. When the processor 302 executes the computer program 303, the computer device can execute any of the aforementioned methods for extracting morphological features of the tongue base region for patients with cervical HPV infection.

[0103] In summary, by analyzing images of tongue retraction (representing tongue contraction) and tongue extension (representing tongue stretching), we can identify and eliminate the characteristic patterns caused by physiological folds to a certain extent, and remove the high-reflectivity features of saliva. This allows us to extract hyperplastic tissue characteristic factors that represent the characteristics of hyperplastic tissue in a relatively objective manner, thereby improving the rationality of hyperplastic tissue feature extraction and assisting in the localization of hyperplastic tissue to a certain extent.

Claims

1. A method for extracting morphological features of the tongue base region in patients with cervical HPV infection, characterized in that, Includes the following steps: Acquire each frame of the target tongue image during the tongue extension and retraction movement of the target patient, and based on the morphology of the tongue region in the target tongue image, filter out the tongue retraction image and the tongue extension image from all the target tongue images; Based on images of tongue retraction and tongue extension, standard retracted tongue root regions and standard extended tongue root regions are constructed. Pixels that match each pixel in the standard extended tongue root region are selected from the standard retracted tongue root region, thus obtaining the matching pixels corresponding to each pixel in the standard retracted tongue root region. The standard retracted tongue root region is divided into blocks to obtain initial blocks, and the target block is selected from all the initial blocks; Physiological wrinkle features are removed based on the similarity between the gradient values ​​of pixels within each target block and their corresponding matching pixels, and the texture stabilization factor for each target block is determined. Saliva specular reflection features are removed based on the texture stabilization factors corresponding to all target blocks, and the co-feature factors corresponding to each target block are determined. Based on the texture stability factor and synergistic feature factor corresponding to all target blocks, construct the proliferative tissue feature factor.

2. The method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 1, characterized in that, The process of selecting images of tongue retraction and tongue extension from all target tongue images based on the morphology of the tongue region in the target tongue image includes: The tongue region is selected from each frame of the target tongue image; The aspect ratio of the minimum bounding rectangle of the tongue region in each frame of the target tongue image is determined as the morphological representative feature factor corresponding to each frame of the target tongue image. The target tongue image with the largest corresponding morphological characteristic factor is selected from all target tongue images and used as the tongue extension image; The target tongue image with the smallest corresponding morphological characteristic factor is selected from all target tongue images and used as the tongue retraction image.

3. The method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 1, characterized in that, The construction of standard retracted tongue root regions and standard extended tongue root regions based on tongue retraction and tongue extension images includes: Based on the geometric center of the tongue region in the tongue retraction image, the initial retracted tongue root region is extracted from the tongue retraction image; Based on the geometric center of the tongue region in the tongue extension image, the initial extended tongue root region is extracted from the tongue extension image; Using a bilinear interpolation algorithm, the initial retracted tongue root region and the initial extended tongue root region are scaled to the same size to obtain the standard retracted tongue root region and the standard extended tongue root region.

4. The method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 3, characterized in that, The step of extracting the initial retracted tongue root region from the tongue retraction image based on the geometric center of the tongue region in the tongue retraction image includes: The main axis of the tongue is obtained by least squares fitting of the contour of the tongue region in the tongue retraction image, and is denoted as the main axis of tongue retraction. The geometric center of the tongue region in the tongue retraction image is translated by a preset offset along the main axis of tongue retraction toward the root of the tongue, and the position point obtained after translation is determined as the target position point; Centered on the target location point, construct a square region with a side length equal to the preset length, and denote it as the initial retraction tongue root region.

5. The method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 1, characterized in that, The step of selecting pixels from the standard extended tongue root region that match each pixel in the standard retracted tongue root region to obtain the matching pixels corresponding to each pixel in the standard retracted tongue root region includes: Pixels with gray values ​​greater than a preset gray threshold are selected from the standard retracted tongue root area and used as retracted highlight points. Pixels with gray values ​​less than or equal to a preset gray threshold are selected from the standard retracted tongue root area and used as non-retracted highlight points. Using the LK optical flow algorithm, pixels that match each non-retracted highlight point are selected from the standard extended tongue root area and used as the matching pixels corresponding to each non-retracted highlight point. Based on the matching pixels corresponding to non-retracted highlight points within the preset neighborhood of each retracted highlight point, the matching pixels corresponding to each retracted highlight point are selected.

6. The method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 5, characterized in that, The step of filtering out matching pixels corresponding to each retracted highlight point based on matching pixels corresponding to non-retracted highlight points within a preset neighborhood of each retracted highlight point includes: Any retracted highlight point is designated as a marked highlight point, and a displacement vector corresponding to each non-retracted highlight point is constructed based on each non-retracted highlight point and its corresponding matching pixel point. The mean value of the displacement vectors corresponding to all non-retracted highlight points in the preset neighborhood of the marked highlight point is determined as the target displacement vector. From the standard extended tongue root region, select the position points whose displacement vector between them and the marked highlight points is equal to the target displacement vector, and use them as reference position points; The pixel closest to the reference position point is determined as the matching pixel corresponding to the marked highlight point.

7. The method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 1, characterized in that, The step of removing physiological wrinkle features based on the similarity between the gradient values ​​of pixels within each target block and their corresponding matching pixels, and determining the texture stabilization factor for each target block, includes: The gradient values ​​corresponding to all pixels within each target block are used to form the original gradient value sequence for each target block. The gradient values ​​of the matching pixels corresponding to all pixels in each target block are used to form the matching gradient value sequence for each target block. The texture stabilization factor for each target block is determined based on the Pearson correlation coefficient between the original gradient value sequence and the matching gradient value sequence corresponding to each target block.

8. The method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 1, characterized in that, The step of removing saliva specular reflection features based on the texture stabilization factors corresponding to all target blocks, and determining the collaborative feature factors corresponding to each target block, includes: Based on the texture stability factor corresponding to the target block, determine the target association value between target blocks; Based on the target association values ​​between target blocks, an organizational spatial association matrix is ​​constructed with the target blocks as nodes, where the target association values ​​are the elements in the organizational spatial association matrix. Based on the organizational spatial correlation matrix, a target degree matrix is ​​constructed, where the cumulative value of each row of elements in the organizational spatial correlation matrix is ​​used as the element on the diagonal of the target degree matrix. Construct a target identity matrix, wherein the target identity matrix, the organization space correlation matrix, and the target degree matrix have the same size; Generate a Laplace matrix based on the target identity matrix, the organization space correlation matrix, and the target degree matrix; Perform eigenvalue decomposition on the Laplacian matrix to obtain multiple eigenvalues ​​and their corresponding eigenvectors; Normalize the eigenvector corresponding to the second smallest eigenvalue obtained by matrix eigenvalue decomposition to obtain the co-eigenvector; The element corresponding to each target block in the collaborative feature vector is determined as the collaborative feature factor for each target block.

9. A method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 8, characterized in that, The step of determining the target association values ​​between target blocks based on the texture stability factor corresponding to the target block includes: Any target block is identified as a marked region. From all target blocks, target blocks whose distance from the marked region is not 0 and whose distance from the marked region is less than or equal to a preset distance threshold are selected and identified as candidate blocks. Each target block other than the candidate blocks in all target blocks is identified as a temporary block, and the target association value between the marked area and each temporary block is set to a preset association value; Based on the difference between the texture stability factor corresponding to the marked region and the texture stability factor corresponding to each candidate block, and the distance between the marked region and each candidate block, the target correlation value between the marked region and each candidate block is determined.

10. A method for extracting morphological features of the tongue base region in patients with cervical HPV infection according to claim 1, characterized in that, The process of selecting the target block from all initial blocks includes: Select the initial blocks whose center point is located within the tongue area from all initial blocks, and use them as the target blocks.

Images