Scar auxiliary examination method based on multivariate fusion analysis of collagen fibers

Abstract

The present invention proposes a scar auxiliary inspection tool based on multivariate fusion analysis of collagen fibers, which is used to fuse and analyze the multivariate structural characteristics of scar tissue collagen fibers, establish a multi-parameter scar auxiliary inspection model to automatically perform scar detection and analysis. The tool can perform multivariate fusion analysis on scar tissue collagen fiber images to obtain multivariate structural parameters such as fiber orientation, fiber curvature, fiber direction variance, and fiber local density. Based on these parameters, a scar detection model is established, which can automatically perform scar detection and analysis. This method can process and analyze collagen fiber images obtained by various optical microscopy techniques, providing multidimensional and mutually complementary collagen fiber structural information, and can more comprehensively analyze the morphological characteristics of collagen fibers and scar tissue. It has the advantages of being fast, highly precise, non-invasive, simple, and highly applicable.

Description

Technical Field

[0001] The present invention belongs to the technical field of biological tissue image analysis and disease detection, and particularly relates to a scar auxiliary inspection method based on collagen fiber multivariate fusion analysis. Background Art

[0002] Scars are a major negative consequence of thermal injury, traumatic injury, and surgery. These scars are hard, painful, itchy, raised, and contracted, leading to localized functional loss and, in some cases, disability. Current treatments for scars include intralesional glucocorticoid injections, laser therapy, surgery, cryotherapy, radiotherapy, and the application of silicone products. Despite the multitude of treatments, clinical results are unsatisfactory. The most important reason is the lack of understanding of the pathophysiology of human scars. Notably, skin scars are unique in humans. Currently, there are no animal models similar to human scars. This further hinders the progress of laboratory research on the pathophysiology of scars, and in turn, the development of clinical applications.

[0003] The disordered arrangement of collagen in scar tissue leads to increased tissue scattering, and the morphological structure of elastin fibers is also significantly different from that of normal skin tissue. While elastin fibers in normal tissue are long, rope-like, spring-like coils and recoils, forming a flexible skin structure, elastin fibers in scar tissue are present in fragmented fibers. The results indicate that the morphology of elastin fibers also undergoes significant changes during scar formation, which can serve as an indicator to identify the boundary between scar tissue and normal skin, assess treatment response, and track scar formation.

[0004] Multiphoton microscopy, based on two-photon excited fluorescence (TPEF) and second harmonic generation (SHG), is one of the most important inventions in the field of bioimaging in recent years. Using an infrared femtosecond laser as an excitation light source, it offers unique advantages, such as reduced sample photobleaching and photodamage, and enhanced penetration depth. It has become a powerful tool for imaging unstained tissue samples and has the potential to noninvasively assess and monitor the morphological structure and functional state of living tissue. Because collagen in the dermis readily generates SHG signals, while elastin more efficiently generates TPEF, multiphoton microscopy can be used for scar detection and is a powerful tool for tracking the progression of scars, guiding treatment plans, and evaluating treatment efficacy.

[0005] Disadvantages of existing technologies:

[0006] H&E (hematoxylin-eosin staining) sections: Currently, doctors generally use H&E sections to judge scar tissue, but relatively speaking, H&E sections consume more manpower and material resources; at the same time, the staining process will destroy cells and is highly invasive; the resolution is low.

[0007] Single parameter analysis: Previously, we proposed to perform single parameter calculation on scar tissue collagen fiber images to characterize and classify them to assist in the inspection. However, due to the single parameter, the classification accuracy is relatively low. Summary of the Invention

[0008] In light of the aforementioned shortcomings of the existing technologies, the present invention proposes a scar-assisted inspection method based on multivariate fusion analysis of collagen fibers. This tool performs multivariate fusion analysis on scar tissue collagen fiber images, obtaining multivariate structural parameters such as fiber orientation, fiber curvature, fiber directional variance, and local fiber density. Based on these parameters, a multi-parameter scar-assisted inspection model is established to automatically perform scar detection and analysis. This tool can process and analyze collagen fiber images obtained using a variety of optical microscopy techniques, offering advantages such as speed, simplicity, accuracy, and wide applicability.

[0009] The present invention is achieved by adopting the following technical solutions:

[0010] The present invention discloses a scar-assisted examination method based on multivariate fusion analysis of collagen fibers. The method is characterized by fusion analysis of the multivariate structural characteristics of scar tissue collagen fibers, and establishing an analysis model for characterizing and calculating scar tissue collagen fibers to assist in scar examination. The construction of the examination tool includes the following steps:

[0011] 1) Using second harmonic generation technology and two-photon fluorescence technology to image collagen fibers in the tissue to be examined, an optical image with micron-level resolution is obtained;

[0012] 2) extracting structural features of collagen fibers from the corresponding collagen fiber image of the tissue to be examined;

[0013] 3) The structural characteristics of collagen fibers were calculated by multivariate structural parameters such as fiber orientation, fiber direction variance, fiber curvature, and fiber local density, and numerical results were obtained;

[0014] 4) marking the location of the second harmonic generation / two-photon fluorescence imaging in the tissue to be examined, then staining the tissue to be examined with hematoxylin-eosin for pathological examination, and then labeling whether the marked area contains scars based on the results of the pathological examination;

[0015] 5) Based on the numerical results of the multivariate structural parameter calculation obtained in step 3) and the label information obtained in step 4), a multi-parameter scar auxiliary examination model is constructed using Fisher discriminant;

[0016] 6) Through training with a large amount of multivariate parameters and label information, a scar auxiliary inspection tool based on collagen fiber multivariate parameter analysis is finally obtained.

[0017] As a further improvement, in step 3) of the present invention, the fiber orientation quantitatively characterizes the spatial orientation of the collagen fibers, providing the ability to quantitatively read the tissue, with a value range of 0°-180°; the fiber curvature quantitatively calculates the curvature of the fibers, characterizing the direction of the collagen fibers in the tissue; the fiber direction variance quantitatively characterizes the degree of order of the collagen fibers, based on directional statistics, and is calculated based on a three-dimensional vector weighted summation algorithm, with a value range of [0,1]. The more disordered the collagen fiber arrangement, the closer the three-dimensional direction variance value is to 1, and the closer the collagen fiber arrangement is to parallel, the closer the three-dimensional direction variance value is to 0; the fiber local density quantifies the spatial distribution of the collagen fibers.

[0018] As a further improvement, the fiber orientation described in the present invention is calculated by a vector weighted summation algorithm, which defines a vector according to the length and intensity changes of the vector in different directions and performs weighted summation on it, and the resolution of the result is at the pixel level; the fiber curvature is obtained by the degree of bending of the collagen fibers in the field, and the resolution of the result is at the pixel level; the fiber direction variance is obtained by the variance of the collagen fiber orientation in the field, and the resolution of the result is at the pixel level; the fiber density is obtained by calculating the ratio of collagen fiber pixels in the field, and the resolution of the result is at the pixel level.

[0019] As a further improvement, in step 5) of the present invention, the calculated fiber direction variance, fiber curvature, and fiber density of the normal tissue image group (class 1), scar image group (class 2), and test sample (class sample) are input into the multi-parameter scar auxiliary inspection model, and Fisher discriminant is used: first, the overall data is read in and stored as matrix w12, and the data corresponding to each class in the matrix is ​​stored as matrices w1, w2, and sample, respectively; the number of samples r1, r2, and r3 of class 1, class 2, and class sample are calculated; then, the mean (matrix) m1 and m2 of class 1 and class 2 are calculated, respectively; the intra-class dispersion matrices s1 and s2 of class 1 and class 2 (covariance matrices, s1 = cov(w1)*(r1-1), s2 = cov(w2)*(r2-1)) are calculated; and the total intra-class dispersion matrix sw (the covariance matrices of class 1 and class 2 are added together) is calculated; Calculate the projection vector w (w = inv(sw)*(m1-m2)': subtract the mean matrices of class 1 and class 2, take the transpose, then multiply by the total intra-class scatter matrix, and finally invert the result) to obtain the Fisher discriminant; calculate the mean y1 and y2 of each class in the one-bit space after projection (y1 = w'*m1', y2 = w'*m2': the mean of class 1 and class 2 is the transpose of the projection vector multiplied by the transpose of class 1 and class 2 respectively); calculate the threshold w0 (class 1 and class 2, w0 = -1 / 2*(y1+y2)); substitute the fiber direction variance, fiber curvature and fiber density in class sample into the discriminant for evaluation and compare with the threshold to determine whether it belongs to class 1 or class 2. If it belongs to class 1, the corresponding output value is 1, and if it belongs to class 2, the corresponding output value is 2; finally, the result is stored in the result single-column matrix.

[0020] As a further improvement, in step 5) of the present invention, the multi-parameter scar-assisted detection model is based on a support vector machine. The calculated structural parameter dataset is divided into a training set and a test set using the leave-one-out method. The training set is used to construct and train the discriminant model, while the test set is used to test the model's discriminative ability for scar detection and optimize the detection model. Multivariate structural parameters such as fiber orientation, fiber curvature, fiber directional variance, and fiber local density are input into the multi-parameter scar-assisted detection model. The complementarity between these multivariate structural parameters ensures the accuracy and reliability of the scar detection model.

[0021] This method utilizes multiphoton microscopy (second harmonic generation and two-photon fluorescence imaging) to analyze scar tissue collagen fibers in a multivariate, fusion-based manner. This method is highly accurate, rapid, non-invasive, and resource-efficient. By calculating multivariate structural parameters such as fiber orientation, fiber curvature, fiber directional variance, and localized fiber density from scar tissue collagen fiber images, it achieves highly accurate, pixel-level resolution analysis of the three-dimensional structure of collagen fibers through multidimensional quantification. This scar-assisted examination tool assists physicians in conducting examinations, further enhancing our understanding of the pathophysiology of human scars.

[0022] Specifically, compared with the existing technical solutions, the present invention has the following advantages:

[0023] 1. This method extracts multi-dimensional structural features from scar tissue collagen fiber images. Compared to existing H&E sectioning methods, it offers the advantages of high precision, rapidity, non-invasiveness, and reduced labor and material resources, while also enabling imaging of small lesions. Furthermore, compared to existing single-structure feature extraction methods, it provides multi-dimensional and complementary collagen fiber structural information, enabling a more comprehensive analysis of the morphological characteristics of collagen fibers and scar tissue.

[0024] 2. Since there is currently no animal model similar to human scars, the scar detection method proposed in this invention, based on multivariate fusion analysis of collagen fibers, has established a scar auxiliary examination tool. By analyzing the structural characteristics of collagen fibers, it provides a more accurate description of the various stages of scar tissue, promotes the medical detection of scar tissue, and thus promotes the progress of laboratory research on scar pathophysiology and the development of clinical applications.

[0025] 3. The structural features used in the present invention, such as fiber orientation, fiber curvature, fiber direction variance and fiber local density, provide quantitative information with pixel-level resolution. By using pseudo-color coding technology, this information is converted into more vivid color images for presentation, with strong information readability, making auxiliary detection simpler and more efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the process for constructing a scar-assisted examination tool for multivariate fusion analysis of collagen fibers;

[0027] Figure 2 This is the optical imaging result of the scar auxiliary examination method based on multivariate fusion analysis of collagen fibers; Figure 2 (a) Multiphoton microscopic images of collagen fibers in normal tissue and scar tissue; Figure 2 (b)-(e) are pseudo-color-coded grayscale images of fiber orientation, fiber direction variance, fiber curvature, and fiber local density; Figure 2(f) Distribution histograms of fiber direction variance, fiber curvature, and fiber local density, respectively.

[0028] Figure 3 This is a schematic diagram of the specific process of the multi-parameter discrimination method - Fisher discrimination;

[0029] Figure 4 This is the calculation result diagram of Fisher discriminant example;

[0030] Figure 5 Schematic diagram of single parameter and multi-parameter classification accuracy. DETAILED DESCRIPTION

[0031] In order to describe the present invention more specifically, the technical solution of the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] The present invention is an auxiliary examination tool for pancreatic cancer based on multivariate parameter analysis of collagen fibers, which integrates and analyzes the multivariate structural characteristics of scar tissue and establishes a scar detection model to assist in scar detection and diagnosis. The specific steps include: Figure 1 As shown:

[0033] Optical imaging: Imaging of collagen fibers in tissues. Imaging tools include second harmonic generation imaging and two-photon fluorescence imaging.

[0034] Calculation of multivariate parameters of collagen fibers: Perform multivariate quantitative analysis on the collagen fiber image of the tissue to be examined obtained in step 1), and calculate the fiber direction variance, fiber curvature and fiber local density respectively.

[0035] Fiber orientation quantitatively characterizes the spatial orientation of collagen fibers and provides the ability to quantitatively read out tissue. In two-dimensional space, the orientation of a fiber is described by the angle θ between it and the x-direction, which ranges from 0° to 180°.

[0036] Fiber curvature quantitatively calculates the spatial curvature of the fiber and characterizes the orientation of collagen fibers in the tissue. The spatial curvature is calculated within the spatial orientation window. First, the difference between the spatial orientation value of all non-center valid pixels in the window and the spatial orientation of the center pixel is calculated N times in total, where N is the number of non-center pixels in the window. This method is used to quantify the change in spatial orientation within the window neighborhood. For fibrous structure data, when the difference is negative or greater than 90°, correction is required. When the difference is negative, its absolute value needs to be taken. When the difference is greater than 90°, its complementary angle needs to be taken. Finally, all the corrected spatial orientation differences are summed up, and the final value is divided by N, and then divided by 90 to obtain the final normalized spatial curvature value. Finally, the image is pseudo-colored according to the calculated spatial curvature value to more clearly represent the spatial curvature value, thereby obtaining the fiber curvature in the image.

[0037] The fiber orientation variance quantitatively characterizes the degree of order of collagen fibers. It is based on directional statistics and is calculated using a vector weighted summation algorithm. To obtain the orientation of the central voxel of an n×n voxel square window in two-dimensional space, the direction of the square's central pixel—defined as the vector passing through the central pixel—is first calculated. The vectors passing through the central pixel are then weighted according to their length and intensity fluctuations along the direction. The direction of the central pixel is defined as the sum of all these weighted vectors. The fiber orientation variance ranges from [0,1]. The more disordered the collagen fibers are, the closer the orientation variance is to 1. The closer the collagen fibers are to parallel alignment, the closer the orientation variance is to 0.

[0038] The local fiber density is calculated by calculating the ratio of collagen fiber pixels within a domain, quantifying the spatial distribution of collagen fibers. As can be seen, compared to scar tissue, normal tissue exhibits smaller fiber orientation variance and more ordered fiber arrangement; smaller fiber curvature and a more parallel arrangement; and a higher local fiber density and a denser fiber arrangement.

[0039] The location of the second harmonic generation / two-photon fluorescence imaging in the tissue to be examined is marked, and then the tissue to be examined is stained with hematoxylin-eosin for pathological examination. Then, based on the results of the pathological examination, the marked area is labeled as to whether it contains scars.

[0040] Based on the numerical results of the multivariate structural parameter calculations and the obtained label information, a support vector machine model was constructed. The presence of scar tissue within the tissue was examined using three parameters derived from the multivariate collagen fiber parameter calculations: fiber orientation variance, fiber curvature, and localized fiber density. This model ensures the complementarity of information provided by the different structural characteristics of collagen fibers, improving accuracy compared to single-parameter analysis and enabling highly accurate scar-assisted examination applications.

[0041] Scar-assisted inspection utilizes a pre-trained support vector machine mathematical model characterized by multivariate parameters of collagen fibers. To train the model, a large number of samples from different scar tissues were imaged and their multivariate parameters were calculated. The multivariate parameter information, along with label information (whether the tissue contains scars), was then used to train the support vector machine model. The detailed steps are as follows: The fiber direction variance, fiber curvature, and fiber density of the normal tissue image group (class 1), scar image group (class 2), and test sample (class sample) were input into the support vector machine model. Fisher discriminant was used: first, the overall data was read in and stored as matrix w12, and the data corresponding to each class in the matrix was stored as matrices w1, w2, and sample, respectively; the number of samples r1, r2, and r3 for class 1, class 2, and class sample were calculated; then the mean (matrix) m1 and m2 for class 1 and class 2 were calculated, respectively; and the intra-class dispersion matrices s1 and s2 (covariance matrix, s1 = cov(w1)*(r1- 1), s2 = cov(w2)*(r2-1)); calculate the total intra-class scatter matrix sw (the covariance matrix of class 1 and class 2 is added); calculate the projection vector w (w = inv(sw)*(m1-m2)': subtract the mean matrix of class 1 from that of class 2, take the transpose, then multiply it with the total intra-class scatter matrix, and finally invert the result) to obtain the Fisher discriminant; calculate the mean y1 and y2 of each class in the projected one-bit space (y1 = w'*m1', y2 = w'*m2 ': The mean of class 1 and class 2 is the transpose of the projection vector and the multiplication of the transpose of class 1 and class 2 respectively); calculate the threshold w0 (class 1 and class 2, w0 = -1 / 2*(y1+y2)); substitute the fiber direction variance, fiber curvature and fiber density in class sample into the discriminant to evaluate and compare with the threshold to determine whether it belongs to class 1 or class 2. If it belongs to class 1, the corresponding output value is 1, and if it belongs to class 2, the corresponding output value is 2; the final result is stored in the result single-column matrix.

[0042] It should be pointed out that although the calculation of collagen fiber multivariate parameters in the present invention involves the calculation of four parameters: fiber orientation, fiber direction variance, fiber curvature and fiber local density, the value of fiber orientation will change due to the change of the reference coordinate system. Therefore, the fiber orientation parameter is not used in the mathematical model used in scar auxiliary examination; however, since the calculation of the fiber direction variance parameter is based on the fiber orientation, its value itself contains the morphological information provided by the fiber orientation, and the value of this parameter is not affected by the spatial coordinate system.

[0043] For a tissue that needs to be examined for scars, the collagen fibers in the tissue are first imaged, and then its various structural parameters are calculated. The parameter values ​​are input into the pancreatic cancer auxiliary examination tool for evaluation and comparison with the trained Fisher discriminant threshold to determine whether it should belong to class 1 or class 2. If it belongs to class 1, the corresponding output value is 1, and if it belongs to class 2, the corresponding output value is 2. That is, an output value of 1 represents a normal tissue image without scars, and an output value of 2 represents a tissue containing scars.

[0044] The specific implementation of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments, but the application of the present invention is not limited thereto.

[0045] Taking specific examples of normal tissue and scar tissue as examples, the steps are as follows: Figure 1 Schematic diagram of the process for constructing a scar-assisted examination tool for multivariate fusion analysis of collagen fibers; Figure 2 This is the optical imaging result of the scar auxiliary examination method based on multivariate fusion analysis of collagen fibers; Figure 2 (a) Multiphoton microscopic images of collagen fibers in normal tissue and scar tissue; Figure 2 (b)-(e) are pseudo-color coded grayscale images of fiber orientation, fiber direction variance, fiber curvature, and fiber local density. 1) First, optical imaging of collagen fibers in scar tissue is performed. Imaging tools include second harmonic generation imaging and two-photon fluorescence imaging. Representative second harmonic generation-two-photon fluorescence images of scar tissue and normal tissue are shown in Figure 1. Figure 2 As shown in (a).

[0046] 2) The resulting collagen fiber image is then subjected to multivariate parameter calculations, including fiber orientation, fiber orientation variance, fiber curvature, and localized fiber density. Fiber orientation quantitatively characterizes the spatial orientation of collagen fibers, providing a quantitative readout of tissue. In two-dimensional space, fiber orientation is described by the angle θ with the x-direction, which ranges from 0° to 180°.

[0047] The fiber orientation variance quantitatively characterizes the degree of order of collagen fibers. It is based on directional statistics and is calculated using a vector weighted summation algorithm. To obtain the orientation of the central voxel of an n×n voxel square window in two-dimensional space, the direction of the square's central pixel—defined as the vector passing through the central pixel—is first calculated. The vectors passing through the central pixel are then weighted according to their length and intensity fluctuations along the direction. The direction of the central pixel is defined as the sum of all these weighted vectors. The fiber orientation variance ranges from [0,1]. The more disordered the collagen fibers are, the closer the orientation variance is to 1. The closer the collagen fibers are to parallel alignment, the closer the orientation variance is to 0.

[0048] Fiber curvature quantitatively calculates the spatial curvature of the fiber and characterizes the orientation of collagen fibers in the tissue. The spatial curvature is calculated within the spatial orientation window. First, the difference between the spatial orientation value of all non-center valid pixels in the window and the spatial orientation of the center pixel is calculated N times in total, where N is the number of non-center pixels in the window. This method is used to quantify the change in spatial orientation within the window neighborhood. For fibrous structure data, when the difference is negative or greater than 90°, correction is required. When the difference is negative, its absolute value needs to be taken. When the difference is greater than 90°, its complementary angle needs to be taken. Finally, all the corrected spatial orientation differences are summed up, and the final value is divided by N, and then divided by 90 to obtain the final normalized spatial curvature value. Finally, the image is pseudo-colored according to the calculated spatial curvature value to more clearly represent the spatial curvature value, thereby obtaining the fiber curvature in the image.

[0049] The local fiber density is calculated by calculating the ratio of collagen fiber pixels within a domain, quantifying the spatial distribution of collagen fibers. As can be seen, compared to scar tissue, normal tissue exhibits smaller fiber orientation variance and more ordered fiber arrangement; smaller fiber curvature and a more parallel arrangement; and a higher local fiber density and a denser fiber arrangement.

[0050] The numerical results of multivariate parameters are as follows Figure 2 In addition to the multivariate parameter results, pseudo-color coded images generated based on the numerical results of different parameters can also be output to more clearly evaluate the structural characteristics of scar tissue. Representative pseudo-color coded images are shown in Figure 2 (be) shown.

[0051] 3) Then, the obtained multivariate parameters are imported into the scar auxiliary inspection tool, evaluated and compared with the trained Fisher discriminant threshold to determine whether it belongs to the normal tissue image group (class 1) or the scar image group (class 2). If it belongs to class 1, the corresponding output value is 1, and if it belongs to class 2, the corresponding output value is 2. The specific calculation and results of each step are as follows: Figure 3 shown. Figure 3Schematic diagram of the specific process of the multi-parameter discrimination method - Fisher discrimination; the multi-parameter discrimination method is to input the calculated fiber direction variance, fiber curvature and fiber density of the normal tissue image group (class 1), scar image group (class 2) and test sample (class sample), and use Fisher discrimination - first read the overall data and save it as matrix w12, and save the data corresponding to each class in the matrix as matrices w1, w2 and sample respectively; calculate the number of samples r1, r2 and r3 of class 1, class 2 and class sample respectively; then calculate the mean (matrix) m1 and m2 of class 1 and class 2 respectively; calculate the intra-class dispersion matrix s1 and s2 of class 1 and class 2 (covariance matrix, s1 = cov(w1)*(r1-1), s2 = cov(w2)*(r2-1)); calculate the total intra-class dispersion matrix sw (the sum of the covariance matrices of class 1 and class 2 ); calculate the projection vector w (w = inv (sw) * (m1-m2) ': subtract the mean matrix of class 1 and class 2 and take the transpose, then multiply it with the total intra-class scatter matrix, and finally perform matrix inversion on the result) to obtain the Fisher discriminant; calculate the mean y1 and y2 of each class in the one-bit space after projection (y1 = w' * m1', y2 = w' * m2': the mean of class 1 and class 2 is the transpose of the projection vector multiplied by the transpose of class 1 and class 2 respectively); calculate the threshold w0 (class 1 and class 2, w0 = -1 / 2 * (y1 + y2)); substitute the fiber direction variance, fiber curvature and fiber density in class sample into the discriminant for evaluation and compare with the threshold to determine whether it should belong to class 1 or class 2. If it belongs to class 1, the corresponding output value is 1, and if it belongs to class 2, the corresponding output value is 2; finally, the result is stored in the result single-column matrix.

[0052] Figure 4 The figure shows the calculation result of the Fisher discriminant example. The 14 groups of data in the sample class are divided into class 1 and class 2 according to the specific values ​​obtained after substituting them into the Fisher discriminant formula and comparing them with the threshold. The output value corresponding to class 1 is 1, and the output value corresponding to class 2 is 2. Figure 4 From the results of this specific example, we can see that 14 groups of data in the class sample were successfully classified, of which 6 were class 1 and 8 were class 2. According to the known classification of the 14 groups of data in the class sample corresponding to the original image, the first 7 groups of data belong to normal tissue (class 1) and the last 7 groups of data belong to scar tissue (class 2). Therefore, only 1 of the 14 groups of data was classified incorrectly, with high classification accuracy.

[0053] Figure 5 Schematic diagram of single parameter and multi-parameter classification accuracy; Figure 5(a)-(c) are schematic diagrams of the results of normal tissue and scar tissue classification using fiber direction variance, curvature and local density as single parameters; Figure 5 (d) Schematic diagram of the results of multi-parameter classification of normal tissue and scar tissue; Figure 5 (e) is the original and cross-validation result table of normal tissue and scar tissue classification using multiple parameters. Figure 5 It can be seen that in this specific example, OCA (original classification accuracy) and CVCA (cross-validation classification accuracy), and the single-parameter classification accuracy (fiber direction variance: 65.0% and 62.5%, curvature: 62.5%, local density: 82.5%) are all lower than the multi-parameter classification accuracy (97.5%). Therefore, multi-parameter can improve the classification accuracy of normal tissue and scar tissue compared with the previous single-parameter method.

[0054] The above examples demonstrate that the present invention, through multivariate structural feature extraction from collagen fiber images of normal tissue and scar tissue, quantitatively characterizes the directional variance, curvature, and local density of collagen fibers. This provides multidimensional, complementary information on collagen fiber structure, enabling a more comprehensive analysis of the morphological characteristics of collagen fibers and scar tissue. Furthermore, pseudo-color images of collagen fiber directional variance, curvature, and local density are provided at pixel-level resolution. This transforms these parameters into vivid, color images for presentation, enhancing readability and making assisted detection simpler and more efficient.

[0055] The above description of the embodiments is intended to facilitate understanding and application of the present invention by those skilled in the art. It will be apparent that those skilled in the art can readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without requiring inventive effort. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention should fall within the scope of protection of the present invention.

Claims

1. A scar auxiliary examination method based on multivariate fusion analysis of collagen fibers, characterized in that The multivariate structural characteristics of scar tissue collagen fibers are integrated and analyzed, and an analytical model for scar tissue collagen fibers is established for characterization and calculation to assist scar examination. The specific steps include: 1) Using second harmonic generation technology and two-photon fluorescence technology to image collagen fibers in the tissue to be examined, an optical image with micron-level resolution is obtained; 2) extracting structural features of collagen fibers from the corresponding collagen fiber image of the tissue to be examined; 3) The structural characteristics of collagen fibers were calculated by multivariate structural parameters such as fiber orientation, fiber direction variance, fiber curvature, and fiber local density, and numerical results were obtained; 4) marking the location of the second harmonic generation / two-photon fluorescence imaging in the tissue to be examined, then staining the tissue to be examined with hematoxylin-eosin for pathological examination, and then marking whether the marked area contains scars based on the results of the pathological examination to obtain label information; 5) Based on the numerical results of the multivariate structural parameter calculation obtained in step 3) and the label information obtained in step 4), a multi-parameter scar auxiliary examination model is constructed using Fisher discriminant; 6) Through training with a large amount of multivariate parameters and label information, a scar auxiliary inspection and analysis model based on collagen fiber multivariate parameter analysis is finally obtained, realizing a scar auxiliary inspection method.

2. The scar auxiliary examination method based on multivariate fusion analysis of collagen fibers according to claim 1, characterized in that: In step 3), the fiber orientation quantitatively characterizes the spatial orientation of the collagen fibers, providing the ability to quantitatively read the tissue, with a value range of 0°-180°; Fiber curvature quantitatively calculates the curvature of the fibers and characterizes the orientation of collagen fibers in the tissue; The fiber direction variance quantitatively characterizes the degree of order of collagen fibers. It is based on directional statistics and calculated using a three-dimensional vector weighted summation algorithm. Its value range is [0,1]. The more disordered the collagen fibers are, the closer the three-dimensional direction variance is to 1. The closer the collagen fibers are to parallel, the closer the three-dimensional direction variance is to 0. The fiber local density quantifies the spatial distribution of collagen fibers.

3. The scar auxiliary examination method based on multivariate fusion analysis of collagen fibers according to claim 2, characterized in that: Fiber orientation is calculated using a vector weighted summation algorithm, which defines vectors based on the changes in their length and intensity in different directions and performs weighted summation on them. The resolution of the result is at the pixel level. Fiber curvature is obtained by the degree of curvature of the collagen fibers within the field, and the resolution of the result is at the pixel level. Fiber direction variance is obtained by the variance of the collagen fiber orientation within the field, and the resolution of the result is at the pixel level. The local fiber density is obtained by calculating the ratio of collagen fiber pixels within the field, and the resolution of the result is at the pixel level.

4. The scar auxiliary examination method based on multivariate fusion analysis of collagen fibers according to claim 1, 2 or 3, characterized in that: In step 5), the calculated fiber direction variance, fiber curvature, and fiber local density of the normal tissue image group, scar image group, and test sample are input into the multi-parameter scar auxiliary examination model, and Fisher discriminant is used: first, the number of samples of the normal tissue image group, scar image group, and test sample is calculated; then, the mean of the normal tissue image group and scar image group is calculated respectively; Calculate the discreteness matrix of the normal tissue image group and the scar image group; Calculate the discreteness matrix of the normal tissue image group and the scar image group; Calculate the projection vectors of the normal tissue image group and the scar image group to obtain the Fisher discriminant; Calculate the mean of the normal tissue image group and the scar image group in the projected one-dimensional space; Calculate the thresholds of the normal tissue image group and the scar image group; substitute the fiber direction variance, fiber curvature and fiber local density in the test sample into the discriminant to evaluate and compare with the threshold to determine whether it belongs to the normal tissue image group or the scar image group. If it belongs to the normal tissue image group, the corresponding output value is 1, and if it belongs to the scar image group, the corresponding output value is 2.

5. The scar auxiliary examination method based on multivariate fusion analysis of collagen fibers according to claim 4, characterized in that: In step 5), the multi-parameter scar auxiliary inspection model is a support vector machine-based model. The calculated structural parameter data set is divided into a training set and a test set according to the leave-one-out method. The training set is used for constructing and training the discriminant model, and the test set is used for testing the model's discriminative ability for scar detection and optimizing the detection model.

Images