Image processing-based fusion traditional chinese medicine treatment of renal fibrosis efficacy evaluation system

Abstract

This invention relates to the field of image processing technology, and more particularly to an image processing-based system for evaluating the efficacy of traditional Chinese medicine in treating renal fibrosis. The system includes: acquiring pathological slide images through an image acquisition and initial screening module, and preliminarily screening target pixels using a preset gradient threshold; extracting multi-level reference pixels from the target pixels using a continuity analysis module to calculate continuity; analyzing the total number of logical feature values ​​with prominent local gradients among the multi-level reference pixels using a comprehensive judgment module, and calculating a comprehensive feature coefficient based on continuity; fusing the comprehensive feature coefficient and gradient magnitude through an adaptive enhancement coefficient mapping module to obtain an adaptive enhancement coefficient; and constructing a two-dimensional enhancement coefficient matrix using the adaptive enhancement coefficient using an image reconstruction and evaluation module, inputting it into a Gaussian inverse masking sharpening filter to complete image reconstruction and achieve efficacy evaluation. This invention eliminates background noise interference and improves the accuracy of pathological efficacy evaluation.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an image processing-based system for evaluating the efficacy of traditional Chinese medicine in treating renal fibrosis. Background Technology

[0002] In the clinical diagnosis and treatment of renal fibrosis and research on traditional Chinese medicine intervention, it is usually necessary to objectively analyze the changes in interstitial collagen deposition and fibrotic areas using renal pathological sections and tissue staining images to assess the degree of pathological improvement before and after traditional Chinese medicine treatment. However, after effective traditional Chinese medicine intervention, the local fibrotic areas of patients will undergo absorption and degradation, showing a gradual weakening of characteristic changes. At the same time, due to the combined effects of physical factors such as differences in tissue section thickness, uneven staining depth, and inherent background noise of imaging equipment, the acquired microscopic pathological images are prone to severe local contrast deficiency, blurred microscopic edges, and interference from background staining particles. Therefore, it is necessary to perform targeted feature enhancement processing on the acquired microscopic images to re-highlight the extremely faint residual lesions after traditional Chinese medicine intervention.

[0003] Traditional image enhancement algorithms commonly used to handle this type of situation, such as Gaussian Unsharp Masking, extract the smooth base of the image through low-pass filtering, obtain the high-frequency detail layer containing edges and noise by subtracting from the original image, and then apply fixed enhancement parameters to the entire high-frequency detail layer to perform indiscriminate magnification before superimposing it back into the original image, thereby improving the visual clarity and edge contrast of the entire image.

[0004] Because the residual weakened lesions caused by the effects of traditional Chinese medicine coexist with the originally clear strong texture of normal tissue and random noise such as free staining residue, when the traditional Gaussian inverse masking sharpening algorithm attempts to amplify the edges of the weakened lesions, it will synchronously and indiscriminately amplify the background staining impurities and thermal noise, resulting in false edge enhancement and breakage of the real weak fiber texture. At the same time, this global amplification will also cause the originally clear strong texture of normal tissue to be severely oversharpened, resulting in visual distortion, which will cover up the real microscopic changes in the efficacy of the medicine and lead to inaccurate evaluation of the efficacy of traditional Chinese medicine by doctors. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an image processing-based system for evaluating the efficacy of traditional Chinese medicine in treating renal fibrosis, in order to solve the technical problem that traditional fixed enhancement parameters cannot adaptively distinguish between weakened lesions and background noise, resulting in false edge enhancement and oversharpening of local tissues.

[0006] This invention provides an image processing-based system for evaluating the efficacy of traditional Chinese medicine in treating renal fibrosis. The system includes the following modules: an image acquisition and initial screening module for acquiring images of renal pathological sections and obtaining gradient amplitude and direction, and screening target pixels based on a preset gradient threshold; a continuity analysis module for calculating the reference degree of each neighboring pixel based on the gradient direction difference and distance between the target pixel and its neighboring pixels, iteratively obtaining multi-level reference pixels by using the neighboring pixel with the highest reference degree, and calculating the continuity of the multi-level reference pixels of the target pixel based on the mean and range of the reference degrees corresponding to the multi-level reference pixels; and a comprehensive judgment module for obtaining the gradient direction of each level of reference pixel. The system identifies directionally adjacent forward and backward neighbor pixels. Based on the gradient magnitude relationship between each reference pixel and its forward and backward neighbor pixels, it extracts the total number of logical feature values ​​with prominent local gradients. The system calculates the comprehensive feature coefficient of the target pixel based on this total number and the continuity. An adaptive enhancement coefficient mapping module calculates correction coefficients to obtain adaptive enhancement coefficients based on the comprehensive feature coefficients and gradient magnitudes of the target pixel. An image reconstruction and evaluation module constructs a two-dimensional enhancement coefficient matrix using the adaptive enhancement coefficients to reconstruct the kidney pathological slide image, performs lesion region segmentation to extract pathological feature parameters, and compares the changing trends of the pathological feature parameters before and after treatment to complete the efficacy evaluation.

[0007] Preferably, the step of selecting target pixels based on a preset gradient threshold includes: acquiring the gradient magnitude of each pixel in the kidney pathological slice image and performing normalization processing to obtain pixels whose normalized gradient magnitude is greater than the preset gradient threshold as target pixels; and pixels whose normalized gradient magnitude is less than or equal to the preset gradient threshold as non-target pixels.

[0008] Preferably, the calculation of the reference degree of each neighboring pixel is specifically calculated using the following formula: In the formula, Indicates the first The first target pixel The degree of reference for each neighboring pixel; Indicates the first The gradient direction of each target pixel; Indicates the first The first target pixel Gradient direction of each neighboring pixel; Indicates the first The target pixel and its first Euclidean distance between neighboring pixels; Represents the standard linear normalization function; This indicates taking the absolute value.

[0009] Preferably, the step of iteratively recursively obtaining multi-level reference pixels by using the neighboring pixel with the highest reference degree includes: using the neighboring pixel with the highest reference degree as the first-level reference pixel of the target pixel; calculating the reference degree between the first-level reference pixel and all its neighboring pixels, and extracting the neighboring pixel with the maximum value as the next-level reference pixel; continuing the iterative recursion with the next-level reference pixel obtained each time as the center, and after reaching a preset number of iterations, obtaining the multi-level reference pixels of the target pixel.

[0010] Preferably, the calculation of the continuity of the multi-level reference pixels of the target pixel is specifically calculated using the following formula: In the formula, Indicates the first The continuity of multi-level reference pixels for each target pixel; Indicates the first The average level of reference for each target pixel across multiple levels of reference pixels; Indicates the first The range of reference levels corresponding to multiple reference pixels for each target pixel.

[0011] Preferably, the extraction of the total number of logical feature values ​​with local gradient prominence includes: if the gradient magnitude of each reference pixel of the target pixel is greater than the gradient magnitude of both the forward neighbor pixel and the reverse neighbor pixel, it is determined that each reference pixel has local gradient prominence, and the logical feature value is 1; otherwise, the logical feature value is 0; the total number of logical feature values ​​with local gradient prominence is the total number of logical feature values ​​of 1.

[0012] Preferably, the calculation of the comprehensive feature coefficient of the target pixel includes: calculating the ratio of the total number of logical feature values ​​with prominent local gradients to the number of multi-level reference pixels; performing standard linear normalization on the continuity of the multi-level reference pixels of the target pixel to obtain the normalized continuity; and calculating the product of the normalized continuity and the ratio to obtain the comprehensive feature coefficient of the target pixel.

[0013] Preferably, the calculation of the correction coefficient to obtain the adaptive enhancement coefficient includes: calculating the first... Correction coefficient for each target pixel , In the formula, Indicates the first The comprehensive feature coefficients of each target pixel; Indicates the first The gradient magnitude of each target pixel; Represents the standard linear normalization function; the preset maximum enhancement coefficient threshold is compared with the normalized i-th... The product of the correction coefficients of the nth target pixel is used as the nth... Adaptive enhancement coefficients for each target pixel.

[0014] Preferably, the reconstruction of the kidney pathology slide image is achieved by constructing a two-dimensional enhancement coefficient matrix using adaptive enhancement coefficients, including: constructing a two-dimensional matrix with the same spatial size as the original kidney pathology slide image; mapping the adaptive enhancement coefficients of the target pixels into the coordinate positions corresponding to the two-dimensional matrix; filling the coordinate positions of non-target pixels with 0, thus obtaining the two-dimensional enhancement coefficient matrix; and inputting the two-dimensional enhancement coefficient matrix into a Gaussian inverse masking sharpening filter to enhance the kidney pathology slide image, thereby obtaining the reconstructed image.

[0015] Preferably, the reconstructed image is input into a U-Net semantic segmentation network with a pre-trained convolutional neural network as the feature extraction backbone, and a fibrotic lesion region mask is output. The proportion of the number of pixels in the mask to the total number of pixels in the image is calculated as the lesion area proportion, and the perimeter of the mask contour is extracted using an edge tracking algorithm and the edge morphology complexity is obtained by combining the roundness formula. The lesion area proportion and edge morphology complexity are compared with those before treatment. If both decrease, the therapeutic effect is judged to be improved; otherwise, the therapeutic effect is judged to be poor.

[0016] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: This invention performs initial screening on original pathological slide images, removing smooth background tissue and non-target pixels such as invalid noise in advance, thus blocking large-area background interference. For target pixels, iterative extraction of the continuity and local gradient prominence distribution of multi-level reference pixels is used to obtain adaptive enhancement coefficients, thereby distinguishing the extremely weak true fibrotic lesions due to the absorption and degradation of traditional Chinese medicine. Image reconstruction and enhancement are completed based on the adaptive enhancement coefficients. Through the processing logic of screening, targeted evaluation, and fusion enhancement, the system avoids the indiscriminate amplification limitation of traditional algorithms. It not only suppresses the pseudo-edge enhancement of random noise and the visual distortion of normal strong textures, but also gives high-magnification enhancement to the remaining weak and stubborn lesions. The reconstructed image output by the system shows the pathological micro-texture, providing a reliable basis for dynamically adjusting the traditional Chinese medicine treatment plan. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a block diagram of the image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis provided in Embodiment 1 of the present invention. Detailed Implementation

[0019] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0020] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0021] This invention provides an image processing-based system for evaluating the efficacy of traditional Chinese medicine treatment for renal fibrosis, such as... Figure 1 As shown, the system includes the following modules: The image acquisition and initial screening module 101 is used to acquire images of kidney pathological sections and screen target pixels.

[0022] It should be noted that in the clinical microscopic scenario of evaluating the efficacy of traditional Chinese medicine intervention in renal fibrosis, the pathological slide images acquired under the microscope contain a large area of ​​background tissue with uniform color, such as normal renal tubular epithelial cell cytoplasm and glomerular capsule fluid. These areas show an extremely smooth transition in the image grayscale space and do not contain structural information reflecting fibrotic collagen deposition or tissue lesions. If feature calculations are performed on all pixels in the entire image, it will not only bring inefficient computing power but also introduce a large amount of background interference errors. Therefore, by extracting the gradient information of the image, the background area is stripped away, and pixels with a certain gradient response are extracted to provide basic data for distinguishing between real texture and discrete noise.

[0023] Specifically, kidney pathology slide images of patients are acquired using medical high-magnification microscopic imaging equipment. The acquired images are then subjected to grayscale dimensionality reduction processing to eliminate color redundancy, obtaining the position information of each pixel. The Sobel operator is used to traverse the grayscale image, calculating the gradient magnitude and gradient direction of each pixel. The gradient magnitudes of all pixels are then normalized to a standard linear normalization, mapped to the 0-1 interval, and a preset gradient threshold is set. All pixels with normalized gradient magnitudes greater than the preset gradient threshold are selected as target pixels; all pixels with normalized gradient magnitudes less than or equal to the preset gradient threshold are selected as non-target pixels.

[0024] It should be added that the preset gradient threshold is used to initially filter and remove flat background and invalid cytoplasmic regions with extremely low gradients in the entire image. The range and selection criteria for its value are as follows: In the early stages, gradient distribution histogram statistical tests were performed on a sample set of hundreds of renal fibrosis pathological slides from different stages of traditional Chinese medicine intervention. The normalized gradient amplitude of the invalid background regions caused by inherent thermal noise from the microscopic imaging equipment and slight uneven adhesion of the staining solution was concentrated below 0.05. Meanwhile, the gradient amplitude response of the weakest residual collagen fiber edges in the later stages of traditional Chinese medicine intervention typically fluctuated between 0.15 and 0.2. By applying different thresholds to the sample set... The acceptance characteristic curve analysis test of the gradient threshold showed that when the gradient threshold was set below 0.05, it would include some background noise, leading to a large number of redundant branches in subsequent calculations; when the gradient threshold was set above 0.2, it would cause excessive removal of weakened lesion features and texture breakage. Therefore, the gradient threshold range was set to [0.05, 0.2], and in this implementation, the value was set to 0.1 to ensure that while removing most of the useless noise background, the early fibrotic edges remaining after the intervention of traditional Chinese medicine were preserved. The implementers can make adaptive adjustments according to the specific staining contrast depth of the actual slide and the inherent noise level requirements of the microscopic imaging equipment in their laboratory.

[0025] At this point, the gradient direction and gradient magnitude of each target pixel have been obtained.

[0026] The continuity analysis module 102 is used to acquire and calculate the continuity of multi-level reference pixels of the target pixel.

[0027] It should be noted that in the microscopic images of renal fibrosis acquired from patients, pixels belonging to the transitional areas of real interstitial collagen deposition, fibrotic stretching texture, and normal tissue structure are physically formed by numerous collagen fiber bundles connected in three-dimensional space. This biological connectivity characteristic, when mapped to a two-dimensional image, manifests as a local spatial continuity in the grayscale gradient changes of the pixels. Conversely, random interference points such as free staining particles during tissue section preparation are spatially discrete and abrupt. Therefore, in order to identify real texture nodes with continuous physical characteristics, it is necessary to find the associated pixels with the closest physical characteristics within the local spatial range of the target pixel, thereby effectively eliminating random noise with disordered orientation.

[0028] Specifically, for any target pixel, taking the target pixel as the center, all pixels within its neighborhood, excluding the target pixel, are obtained as the neighboring pixels of the target pixel. In this embodiment, the neighborhood range adopts a 5×5 local pixel matrix neighborhood, which can just cover the reasonable extension range of the cross-section of a single microscopic collagen fiber bundle under the high magnification microscope field of view. This can ensure the diversity of optimization directions and avoid introducing irrelevant background interference from too far away.

[0029] Based on the difference in gradient direction and spatial distance between the target pixel and its neighboring pixels, the reference level of each neighboring pixel is calculated. The specific calculation formula is as follows: In the formula, Indicates the first The first target pixel The degree of reference for each neighboring pixel; Indicates the first The gradient direction of each target pixel; Indicates the first The first target pixel Gradient direction of each neighboring pixel; Indicates the first The target pixel and its first Euclidean distance between neighboring pixels; Represents the standard linear normalization function; This indicates taking the absolute value.

[0030] in, Reflects the first The target pixel and its first Coordination of neighboring pixels in the gradient direction The smaller the value, the better. The target pixel and its first The high degree of consistency in the grayscale variation trends of neighboring pixels indicates that they may physically belong to the same collagen fiber bundle. The larger; Reflects the first The target pixel and its first The compactness of the spatial physical distribution of the nth neighboring pixels; the larger this value, the more compact the spatial distribution of the nth neighboring pixels. The target pixel and its first The closer the distance between the neighboring pixels, the better; in summary, only when the first... The target pixel and its first When the orientation and height of neighboring pixels are consistent and their spatial distance is extremely close Only then will it approach the maximum value; the larger this value, the higher its rank. The more neighboring pixels a node is, the more qualified it is to be the next level extension node of the real texture.

[0031] Furthermore, simply obtaining the reference level of a single step within a local neighborhood can only prove that the pixel in that neighborhood has the possibility of local extension over a very short distance, but cannot confirm that it is a complete tissue fiber texture. In order to verify whether this tiny extension has a long-range physically stable structure, it is necessary to simulate the real growth path of collagen fiber bundles in the tissue and perform multi-level tracking based on the local correlation point with the highest confidence, and then measure its long-range continuity from the mean performance and fluctuation variance of the entire path.

[0032] Therefore, the reference degree of all neighboring pixels of the target pixel is traversed, and the neighboring pixel corresponding to the maximum reference degree is selected as the first-level reference pixel of the target pixel. With the first-level reference pixel as the center, the above-mentioned acquisition of neighboring pixels and calculation of reference degree are continued to iterate to find the next level reference pixel until the preset number of iterations is met and the iteration stops, thus obtaining the multi-level reference pixels and corresponding reference degree of the target pixel.

[0033] It should be added that, based on the pixel-scale measurement tests of a large number of high-magnification microscopic pathological sections in the early stage, the random noise caused by staining impurities and other factors has extremely weak continuity. The short lines that are occasionally connected are generally less than 8 pixels in length. Therefore, this embodiment selects 8 times as the number of iterations, which is in line with the lower limit of the physical scale of real tissue texture, and achieves the best detection balance between effectively filtering discrete noise and completely tracking weak lesions.

[0034] The continuity of the multi-level reference pixels of the target pixel is calculated based on the mean and range of the reference degree corresponding to the multi-level reference pixels of the target pixel. The specific calculation formula is as follows: In the formula, Indicates the first The continuity of multi-level reference pixels for each target pixel; Indicates the first The average level of reference for each target pixel across multiple levels of reference pixels; Indicates the first The range of reference levels corresponding to multiple reference pixels for each target pixel.

[0035] in, Reflecting the The average optimization ability of multi-level reference pixels of a target pixel in the entire spatial recursive expansion process. The larger this value is, the stronger the directional and spatial correlation between the nodes on the entire path. As a stability penalty coefficient in the path extension process, through To capture fluctuation defects in the extended path, if the reference level of a certain level reference pixel undergoes a drastic jump or abrupt change during the extension of the target pixel along a multi-level pixel sequence path, then the range value... It will increase rapidly, which will lead to the product term Significantly reduced; combining the two, when The larger the value, the more likely the target pixel is to maintain a very high degree of correlation in the path extension homogeneity, and the entire extension process is free of any discontinuities or abrupt changes. This means that the target pixel is attached to a real, long-range, and physically continuous histopathological structure.

[0036] At this point, the continuity of multi-level reference pixels for each target pixel is obtained.

[0037] The comprehensive determination module 103 is used to calculate the comprehensive feature coefficients of the target pixel.

[0038] It should be noted that in the microscopic imaging of pathological sections, pixels with spatial continuity are obtained through the above steps. However, in some smooth transition areas caused by uneven tissue section thickness or local accumulation of staining solution, the grayscale gradient also has continuity. These areas are meaningless for evaluating the efficacy of fibrosis treatment. From a pathological morphology perspective, no matter how thin and weak the collagen fiber bundles become after traditional Chinese medicine intervention, as an independent filamentous or ribbon-like physical structure, the grayscale change at its location is more dramatic than that of the adjacent matrix areas on both sides perpendicular to its direction, like a slightly raised ridge on a plain, manifesting as a peak in a local area. Therefore, it is necessary to perform transverse cross-sectional analysis of each level of multi-level reference pixels along its gradient direction. By determining whether it has the local gradient prominence feature like a ridge, the true center point of the lesion texture that needs to be evaluated by the doctor can be extracted from the meaningless smooth gradient area.

[0039] Specifically, for any target pixel with multiple levels of reference pixels, each level of reference pixel is traversed along its gradient direction, i.e., the normal direction perpendicular to the edge or texture direction, to obtain the positive and negative neighbor pixels adjacent to the reference pixel. Based on the gradient magnitude of the reference pixel and its adjacent positive and negative neighbor pixels, it is determined whether the reference pixel possesses local gradient prominence features. The specific calculation formula is as follows: In the formula, Indicates the first The first target pixel The reference pixels at each level possess logical feature values ​​that highlight local gradients; Indicates the first The first target pixel The gradient magnitude of the reference pixel; and They represent along the first The first target pixel The gradient magnitudes of the forward and reverse neighboring pixels in the gradient direction of the reference pixel.

[0040] Among them, the The first target pixel The gradient magnitude of the reference pixel must be greater than the gradient magnitudes of its two neighboring pixels along the normal direction of the cross section. When both conditions are met... The value is 1, which proves the first... The first-level reference pixel is at the extreme peak of grayscale variation, meaning that the first... Each target pixel may belong to the skeletal response point of the lesion edge or fibrotic texture.

[0041] Furthermore, considering that isolated noise points may also form peaks in their tiny local areas, the real fiber texture is not only a local peak itself, but the nodes on the entire continuous sequence path it constitutes should also generally have this ridge characteristic. Therefore, it is necessary to statistically analyze the proportion of multi-level reference pixels of the target pixel that satisfy this prominent feature, and fuse it with the aforementioned continuous features to comprehensively assess the confidence that the target pixel belongs to the real pathological edge structure.

[0042] Specifically, for any target pixel, the logical feature values ​​of the multi-level reference pixels with prominent local gradients are traversed, and the total number of logical feature values ​​that are 1 is counted. Based on the proportion of the total number and the continuity of the multi-level reference pixels of the target pixel, the comprehensive feature coefficient of the target pixel is calculated, and the specific calculation formula is as follows: In the formula, Indicates the first The comprehensive feature coefficients of each target pixel; Indicates the first The total number of logical feature values ​​of 1 with prominent local gradients among the multi-level reference pixels of a target pixel; Indicates the number of multi-level reference pixels; Indicates the first The continuity of multi-level reference pixels for each target pixel; This represents the standard linear normalization function.

[0043] in, Reflects the first The distribution density of local extrema among multi-level reference pixels of a target pixel; the larger this ratio, the more it indicates that this continuous path is not a gradual transition caused by uneven lighting, but rather an organizational texture skeleton line; compare it with the normalized... The continuous multiplication of multi-level reference pixels for each target pixel achieves complementary physical properties: ensuring both the first and second target pixels are identical. The target pixel structure possesses long-range stable extensibility in the longitudinal direction of space, while also ensuring significant structural sharpness in the cross-section, ultimately resulting in... The larger it is, the more likely it is to be the first The more likely the target pixel is to be an objectively existing and structurally significant pathological tissue feature, the higher the probability that the first pixel is a pathological tissue feature. The confidence level of using these target pixels as a reference for doctors to subsequently determine the efficacy of traditional Chinese medicine interventions is extremely high.

[0044] At this point, the comprehensive feature coefficients of each target pixel have been obtained.

[0045] The enhancement coefficient adaptive mapping module 104 is used to obtain the adaptive enhancement coefficient of the target pixel.

[0046] It should be noted that traditional Gaussian anti-masking sharpening algorithms typically use a globally uniform fixed constant as the enhancement coefficient. However, in the scenario of evaluating traditional Chinese medicine treatment for renal fibrosis, using a globally fixed coefficient would not only severely oversharpen the originally clear normal tissue texture, leading to distortion, but would also amplify free staining noise equally, thereby obscuring the faint real lesions that gradually dissipate and weaken after traditional Chinese medicine intervention. Therefore, by changing the parameter input logic of the traditional algorithm, the fixed enhancement coefficient is adaptively adjusted by combining the comprehensive feature coefficient and gradient magnitude of each target pixel, ensuring that the magnification is higher in areas that may belong to real continuous lesions and are more blurred and weak in visual display; conversely, strong textures or pure isolated noise are suppressed.

[0047] Specifically, based on the comprehensive feature coefficients and gradient magnitude of any target pixel, the correction coefficient for that target pixel is calculated, as follows: In the formula, Indicates the first Correction coefficients for each target pixel; Indicates the first The comprehensive feature coefficients of each target pixel; Indicates the first The gradient magnitude of each target pixel; This represents the standard linear normalization function.

[0048] Specifically, by squaring the comprehensive feature coefficients of the target pixels, a weight discontinuity is created in a non-linear manner between the high-confidence real fiber texture and ordinary random noise. The larger the value, the more significant the effect. The higher the confidence level of a target pixel as the boundary of a real lesion, the better. From a clinical pathological perspective, effective intervention with traditional Chinese medicine leads to the gradual absorption and degradation of collagen in the fibrotic area, causing the physical hard boundary of the lesion to dissipate. This pathological dissipation and degradation, when mapped to the grayscale space of a two-dimensional image, manifests as a sharp decrease in the rate of grayscale change between adjacent pixels in the local area, i.e., a significant reduction in the original gradient amplitude. It becomes extremely small, therefore, The smaller the value, the more blurred and weak the lesion at the target pixel will appear visually after being affected by the drug effect; A targeted compensation logic was then constructed: the correction coefficient is only obtained when the target pixel is highly identified as a real continuous lesion and the current pathological visual features are extremely weak. That's when it reaches its peak.

[0049] Furthermore, a maximum enhancement coefficient threshold is preset, and the correction coefficient of the target pixel is normalized to obtain the normalized correction coefficient of the target pixel; the product of the preset maximum enhancement coefficient threshold and the normalized correction coefficient of any target pixel is used as the adaptive enhancement coefficient of the target pixel.

[0050] Among them, the maximum enhancement coefficient threshold is used to control the maximum enhancement ratio that the anti-mask sharpening filter can apply to the local high-frequency details of the target pixel. The value range is usually [2,5]. After verification with a large number of data samples in the early stage, the preferred value of 3 is taken in this embodiment to ensure that while highlighting the weak collagen fiber characteristics after the intervention of traditional Chinese medicine to the greatest extent, it will not cause excessive overflow of local pixel gray values ​​and visual distortion of pathological morphology. The implementer can make flexible adaptive adjustments according to the specific needs of the original staining contrast depth of the actual pathological slide and the overall photoelectric conversion quality of the sensor of the microscopic imaging equipment.

[0051] At this point, the adaptive enhancement coefficient for each target pixel has been obtained.

[0052] The image reconstruction and evaluation module 105 is used to construct a two-dimensional enhancement coefficient matrix to reconstruct kidney pathological slide images, and to perform lesion region segmentation to extract pathological feature parameters, analyze the changing trends of pathological feature parameters before treatment to complete efficacy evaluation.

[0053] The proportion of residual fibrotic pixels is extracted and compared in the reconstructed image to complete the efficacy assessment.

[0054] It should be noted that this step replaces the global fixed coefficients in the traditional Gaussian anti-mask sharpening algorithm with the adaptive enhancement coefficients of the target pixels extracted above, which can produce a significant differentiation enhancement effect: that is, pure random noise is not amplified, normal strong edges are kept in their original state to avoid over-sharpening, and lesion areas that have undergone effective intervention by traditional Chinese medicine and collagen degradation leading to weakened features are enhanced by the highest factor, so that they can be clearly seen again in the microscopic field of view.

[0055] Specifically, a two-dimensional matrix with the same spatial size as the original kidney pathology slide image is constructed. The adaptive enhancement coefficients of the target pixels are mapped and filled into the corresponding coordinate positions of the two-dimensional matrix. For non-target pixels, 0 is directly filled into the corresponding coordinate positions of the two-dimensional matrix by default, resulting in a two-dimensional enhancement coefficient matrix. The two-dimensional enhancement coefficient matrix is ​​then mapped point by point and substituted into the reconstruction equation of the Gaussian inverse masking sharpening filter. By extracting the high-frequency detail layer of the original pathology image and applying the two-dimensional enhancement coefficient matrix for pixel-by-pixel multiplication and local scaling, the image is finally superimposed with the low-frequency smoothing layer to complete the image reconstruction.

[0056] Subsequently, the adaptively enhanced and reconstructed images were input into the system for efficacy evaluation. The processing procedure is as follows: The system acquires a pre-trained convolutional neural network and a U-Net semantic segmentation network. The reconstructed image is input into the pre-trained convolutional neural network for deep feature extraction. Combined with the U-Net semantic segmentation network model, the fibrotic regions in the image are segmented, and a fibrotic lesion mask is output. Based on the lesion mask obtained from the above segmentation, the system automatically extracts multi-dimensional pathological feature parameters, mainly including: lesion area ratio and edge morphology complexity. The lesion area ratio is the proportion of the number of pixels in the lesion mask to the total number of pixels in the image. The edge morphology complexity is obtained by extracting the perimeter of the lesion contour using an edge tracking algorithm and calculating the edge morphology complexity using a circularity calculation formula.

[0057] By comparing the aforementioned multidimensional pathological feature parameters with the pathological feature parameters extracted by the system before the patient received traditional Chinese medicine treatment, if the proportion of lesion area decreases and the complexity of edge morphology decreases, it indicates that the current degree of renal fibrosis has been effectively absorbed and improved under the intervention of traditional Chinese medicine; otherwise, it indicates that the fibrosis has not been effectively controlled and the treatment effect is deemed poor.

[0058] It should be noted that, in order to achieve high-precision lesion segmentation, the system pre-trained the model. The specific pre-training process is as follows: a large number of historical kidney pathology slide image samples are acquired, and professional pathologists manually annotate the real fibrotic lesion areas in the images at the pixel level to generate corresponding real binary label masks to construct a training sample set; the training sample set is input into the initial convolutional neural network and U-Net semantic segmentation network in batches to obtain the predicted segmentation masks; the error between the predicted mask and the real label mask is calculated using a preset loss function, and based on this error, an optimization algorithm is used for backpropagation to iteratively update the weight parameters in the network model layer by layer until the model converges, thereby obtaining the pre-trained neural network model.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A system for evaluating the efficacy of traditional Chinese medicine treatment for renal fibrosis based on image processing, characterized in that: The system includes the following modules: The image acquisition and initial screening module is used to acquire images of kidney pathological sections and obtain gradient magnitude and gradient direction, and to screen target pixels based on preset gradient thresholds; The continuity analysis module is used to calculate the reference degree of each neighboring pixel based on the gradient direction difference and distance between the target pixel and its neighboring pixels. It iterates and recursively with the neighboring pixel with the highest reference degree to obtain multi-level reference pixels. Based on the mean and range of the reference degree corresponding to the multi-level reference pixels, the continuity of the multi-level reference pixels of the target pixel is calculated. The comprehensive determination module is used to obtain the forward neighbor pixels and backward neighbor pixels of each reference pixel along its gradient direction; extract the total number of logical feature values ​​with local gradient prominence based on the relationship between the gradient magnitude of each reference pixel and the forward neighbor pixels and backward neighbor pixels; and calculate the comprehensive feature coefficient of the target pixel based on the total number and the continuity. The adaptive enhancement coefficient mapping module is used to calculate the correction coefficients based on the comprehensive feature coefficients and gradient magnitude of the target pixel to obtain the adaptive enhancement coefficients. The image reconstruction and evaluation module is used to construct a two-dimensional enhancement coefficient matrix through adaptive enhancement coefficients to reconstruct kidney pathological slide images, and to perform lesion region segmentation to extract pathological feature parameters. The module compares the changes in pathological feature parameters with those before treatment to complete the efficacy evaluation.

2. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The process of filtering target pixels based on a preset gradient threshold includes: The gradient magnitude of each pixel in the kidney pathological section image is obtained and normalized. Pixels with normalized gradient magnitude greater than a preset gradient threshold are selected as target pixels; pixels with normalized gradient magnitude less than or equal to the preset gradient threshold are selected as non-target pixels.

3. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The specific formula for calculating the reference level of each neighboring pixel is as follows: ； In the formula, Indicates the first The first target pixel The degree of reference for each neighboring pixel; Indicates the first The gradient direction of each target pixel; Indicates the first The first target pixel Gradient direction of each neighboring pixel; Indicates the first The target pixel and its first Euclidean distance between neighboring pixels; Represents the standard linear normalization function; This indicates taking the absolute value.

4. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The process of iteratively obtaining multi-level reference pixels by using the neighboring pixels with the highest reference degree includes: The first-level reference pixel of the target pixel is selected from the neighboring pixels with the highest reference degree. The reference degree between the first-level reference pixel and all its neighboring pixels is calculated, and the neighboring pixel with the highest value is extracted as the next-level reference pixel. The iteration is continued with the next-level reference pixel obtained each time as the center. After reaching the preset number of iterations, the multi-level reference pixels of the target pixel are obtained.

5. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The specific formula for calculating the continuity of the multi-level reference pixels of the target pixel is as follows: ； In the formula, Indicates the first The continuity of multi-level reference pixels for each target pixel; Indicates the first The average level of reference for each target pixel across multiple levels of reference pixels; Indicates the first The range of reference levels corresponding to multiple reference pixels for each target pixel.

6. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The total number of logical feature values ​​with prominent local gradients extracted includes: If the gradient magnitude of each reference pixel at the target pixel level is greater than the gradient magnitude of both the forward neighbor pixel and the reverse neighbor pixel, then each reference pixel at the target pixel level is determined to have local gradient prominence, and its logical feature value is set to 1; otherwise, the logical feature value is set to 0; the total number of logical feature values ​​with local gradient prominence is the total number of logical feature values ​​set to 1.

7. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The calculation of the comprehensive feature coefficients of the target pixel includes: Calculate the ratio of the total number of logical feature values ​​with prominent local gradients to the number of multi-level reference pixels; perform standard linear normalization on the continuity of the multi-level reference pixels of the target pixel to obtain the normalized continuity; calculate the product of the normalized continuity and the ratio to obtain the comprehensive feature coefficient of the target pixel.

8. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The calculation of the correction coefficients to obtain the adaptive enhancement coefficients includes: Calculate the first Correction coefficient for each target pixel , In the formula, Indicates the first The comprehensive feature coefficients of each target pixel; Indicates the first The gradient magnitude of each target pixel; Represents the standard linear normalization function; the preset maximum enhancement coefficient threshold is compared with the normalized i-th... The product of the correction coefficients of the nth target pixel is used as the nth... Adaptive enhancement coefficients for each target pixel.

9. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 2, characterized in that, The method of constructing a two-dimensional enhancement coefficient matrix using adaptive enhancement coefficients to reconstruct kidney pathology slide images includes: A two-dimensional matrix with the same spatial size as the original kidney pathology slide image is constructed. The adaptive enhancement coefficients of the target pixels are mapped and filled into the coordinate positions corresponding to the two-dimensional matrix. For non-target pixels, 0 is filled into the coordinate positions corresponding to the two-dimensional matrix to obtain the two-dimensional enhancement coefficient matrix. The two-dimensional enhancement coefficient matrix is ​​then input into a Gaussian inverse masking sharpening filter to enhance the kidney pathology slide image, resulting in the reconstructed image.

10. The image processing-based fusion traditional Chinese medicine treatment efficacy evaluation system for renal fibrosis according to claim 1, characterized in that, The process of segmenting the lesion region to extract pathological feature parameters and comparing the changes in these parameters with those before treatment to complete the efficacy evaluation includes: The reconstructed image is input into a U-Net semantic segmentation network with a pre-trained convolutional neural network as the backbone for feature extraction, and the network outputs a mask of the fibrotic lesion region. The proportion of the number of pixels in the mask to the total number of pixels in the image is calculated as the lesion area ratio, and the perimeter of the mask contour is extracted using an edge tracking algorithm and the edge morphology complexity is obtained by combining it with the roundness formula. The lesion area ratio and edge morphology complexity are compared with those before treatment. If both decrease, the therapeutic effect is judged to be improved; otherwise, the therapeutic effect is judged to be poor.

Images