Breast cancer examination tool based on collagen fiber multi-element fusion analysis

Abstract

The application discloses a breast cancer examination tool based on collagen fiber multi-element fusion analysis, which is used for quantitative analysis of breast tissue collagen fiber structure characteristics and examination of breast tissue cancer level. The application can multidimensionally and comprehensively extract structure information of collagen fibers, and through multi-element fusion analysis of multiple structure parameters, the application improves information complementarity of breast tissue morphological characteristic analysis; the application realizes visualization of collagen fiber structure characteristics by using a pixel level resolution quantitative method and combining with a pseudo-color coding technology, and enhances intuitiveness of information; the application classifies and analyzes three potential biomarkers representing different breast cancer levels, provides a high-precision, automatic and rapid breast cancer level examination method, and has greater application potential compared with a traditional breast cancer examination method.

Description

Technical Field

[0001] This invention belongs to the field of image analysis and disease examination technology of biological tissues, and specifically relates to a breast cancer examination tool based on collagen fiber multivariate fusion analysis. Background Technology

[0002] Breast cancer is the most common malignant tumor in women, posing a significant threat to women's health and survival. "Early diagnosis and early treatment" are the most effective means to reduce breast cancer mortality and improve patients' quality of life. Currently, the main method for examining breast cancer is histopathology, which requires a series of processes including sampling, fixation, paraffin embedding, sectioning, dewaxing, and staining. This method has several drawbacks: First, the examination process is complex and time-consuming, requiring significant manpower and resources; second, the introduction of staining agents that damage tissue structure can affect the accuracy of the results; third, the results rely on the doctor's experience, and their reliability is inevitably affected by subjective factors.

[0003] The extracellular matrix (ECM) is a complex network composed of various macromolecules with unique biomechanical and chemical mechanisms, playing a crucial role in many life processes. Collagen is the main protein in various connective tissues of animals, existing in the form of long, thin fibrils. Tough collagen fiber bundles are a major component of the ECM, providing support for cells in most tissues. Multiple studies have shown that changes in the mechanical and biochemical state of collagen fibers are associated with tumor evolution. Furthermore, the spatial structure of collagen fibers provides potential biomarkers for detecting diseased tissues, assessing treatment response, locating damage, or detecting the development of engineered tissues.

[0004] In studies on collagen fibers in breast tissue, researchers have defined three tumor-associated collagen signals (TACS1-3) that can help identify cancer levels. TACS1 is defined by collagen fibers that curve around the tumor and serves as a marker for locating small tumors. TACS2 is defined by stretched collagen fibers, resulting from tumor growth causing collagen fibers to align parallel to the tumor boundary. TACS3 is defined by collagen fibers that are irregularly aligned perpendicular to the tumor boundary; this state of collagen fibers promotes tumor migration. TACS are closely related to collagen matrix remodeling and tumor development and invasion, and are potential biomarkers for predicting breast cancer levels. Quantitative characterization and classification of TACS are of great significance for understanding the structure-function relationship and cancer level of breast tissue.

[0005] Currently, methods for quantifying the spatial structure of collagen fibers and related breast cancer screening tools are scarce and have certain limitations. First, these tools can only quantify single structural features of collagen fibers, lacking techniques for fusion analysis of multiple structural features. Analysis techniques based on single structural features, while effective for disease detection, have lower accuracy. Second, the morphological information provided by these tools is insufficient. Quantifying the structural features of collagen fibers from multiple perspectives can provide effective information for analyzing the cancer level of breast tissue. Third, there is a lack of analytical methods to finely correlate changes in the structural features of collagen fibers with the cancer level of breast tissue, thus failing to accurately analyze the cancer level of the tissue. Fourth, these tools primarily analyze whole-tissue images and cannot reflect localized cancer characteristics. In summary, there is an urgent need for a multi-dimensional quantitative analysis tool for collagen fibers to perform pixel-level resolution multivariate fusion analysis of extracellular matrix images of breast tissue, and to combine this with three tumor-related collagen signal information to analyze the cancer level, thereby enabling early detection of small breast cancer lesions. Summary of the Invention

[0006] In view of the above, this invention proposes a breast cancer screening tool based on multivariate fusion analysis of collagen fibers. This system and method can perform multivariate fusion analysis on collagen fiber images of breast tissue to obtain multivariate structural parameters such as fiber orientation, fiber direction variance, fiber density, and cell-fiber relative angles. Based on these parameters, a breast cancer level discrimination model is established to assist in breast cancer screening. This method can process and analyze collagen fiber images obtained from various optical microscopy techniques, and has the advantages of being fast, simple, accurate, and highly applicable.

[0007] This invention is achieved using the following technical solution:

[0008] This invention discloses a breast cancer screening tool based on collagen fiber multivariate fusion analysis. It integrates and analyzes the multivariate structural features of breast tissue to establish a breast cancer level discrimination model to assist in breast cancer screening. The screening tool includes the following usage methods:

[0009] 1) Image the collagen fibers in the extracellular matrix of the breast tissue to be examined using two-photon fluorescence and second harmonic imaging techniques to obtain optical images with micron-level resolution;

[0010] 2) Extract the multi-dimensional structural features of collagen fibers from optical images of breast tissue to obtain quantitative results including fiber orientation, fiber direction variance, cell-fiber relative angle, and fiber density;

[0011] 3) Hematoxylin-eosin staining of breast tissue is used for cancer pathological examination. The cancer level of the two-photon fluorescence / second harmonic image obtained in step 1) is marked according to the examination results.

[0012] 4) Based on the multivariate structural feature quantization results calculated in step 2) and the image cancer level labeling obtained in step 3), a support vector machine model is established;

[0013] 5) In the support vector machine model, the quantization result of a sample based on its multivariate structural features is defined as a vector. The quantification results of the multi-dimensional structural features include fiber orientation variance, cell-fiber relative angle, and fiber density, which are respectively derived from v i ,r i ,c i This indicates that the sample vector

[0014] In the model, a separating hyperplane is found in the sample space to separate breast tissue samples of two different cancer levels. The equation of the separating hyperplane is:

[0015]

[0016] in Let b be the normal vector of the segmenting hyperplane, and let b be the displacement of the segmenting hyperplane. The following discriminant is used to determine the cancer level of the breast tissue:

[0017]

[0018] Where y i It is the cancer lesion marker information obtained in step 2), y i =1 indicates that the image belongs to type 1, y i =2 indicates that the image belongs to type 2;

[0019] (6) The model is trained using the dataset of image cancer level markers and multivariate structural feature quantification results. The optimal segmentation hyperplane is found by solving the following equation to obtain the values ​​of ω and b:

[0020]

[0021] Where st indicates that the conditions must be met simultaneously, and m is the total number of samples;

[0022] 7) Construct the optimal segmentation hyperplane based on the obtained ω and b values ​​to distinguish collagen fiber samples at two different cancer levels, and establish a binary classifier for breast cancer level;

[0023] 8) The actual collagen fiber sample cancer level type is divided into three categories. Based on the training method in steps 5)-7), three breast cancer level binary classifiers are constructed. Each classifier is used to distinguish a certain class from all other classes. When testing collagen fiber samples with unknown cancer level, it is input into each classification function. The class with the largest function value is determined to be the cancer level type of the unknown collagen fiber sample, thus obtaining a breast cancer screening tool based on collagen fiber multivariate fusion analysis.

[0024] As a further improvement, in step 3) of the present invention, the fiber orientation is obtained by calculating the difference between the spatial orientation of each collagen fiber pixel and its surrounding collagen fibers, and the resolution of the obtained fiber orientation quantification result is at the pixel level; the fiber direction variance is obtained by calculating the variance of the spatial orientation of each collagen fiber pixel and its surrounding collagen fiber pixels, and the resolution of the obtained fiber direction variance quantification result is at the pixel level; the cell-fiber relative angle is obtained by calculating the difference between the spatial orientation of each collagen fiber pixel and its nearest neighbor cell boundary pixel, and the resolution of the obtained cell-fiber relative angle quantification result is at the pixel level; the fiber density is obtained by calculating the proportion of collagen fiber pixels in the neighborhood, and the resolution of the obtained fiber density quantification result is at the pixel level.

[0025] As a further improvement, the fiber orientation of this invention is used to quantify the spatial arrangement direction of collagen fibers, with a value range of 0-180°; the fiber orientation variance is used to describe the degree of orderliness of the collagen fiber arrangement, with a value range of 0-1, where a value closer to 0 indicates a more parallel and orderly arrangement of collagen fibers, and a value closer to 1 indicates a more scattered and disordered arrangement of collagen fibers; the cell-fiber relative angle is used to quantify the angular difference between the collagen fiber and the central cell boundary, with a value range of 0-90°, where a value of 0° indicates that the collagen fiber and the cell boundary are completely parallel, and a value of 90° indicates that the collagen fiber and the cell boundary are perpendicular; the fiber density is used to quantify the spatial distribution of collagen fibers, with a value range of 0-1, where a value closer to 0 indicates a lower collagen fiber content, and a value closer to 1 indicates a higher collagen fiber content.

[0026] As a further improvement, the segmentation hyperplane normal vectors of the three breast cancer level binary classifiers obtained in steps 5)-8) of this invention are... The values ​​are [0.37589792, 1.01374887, -0.34803396], [-1.33374423, -0.64543886, 2.45260154], and [0.91955191, -0.5849421, -1.98512522], respectively, and the displacements b of the hyperplane are 0.30250453, 0.36193904, and 0.16197132, respectively.

[0027] Compared with existing technical solutions, the present invention has the following advantages:

[0028] 1. Compared with traditional histopathological methods, this invention does not require complicated section preparation, saving manpower and resources, and does not require the introduction of external staining agents, enabling non-destructive, real-time monitoring and evaluation of tissues.

[0029] 2. Compared with traditional histopathological methods, this invention can obtain examination results with specific numerical values ​​to objectively evaluate the cancer level of breast tissue, thus enhancing the reliability of the examination results.

[0030] 3. Compared to traditional collagen fiber analysis tools that can only quantify single structural features of collagen fibers, this invention can quantify multiple structural features of collagen fibers, providing multi-dimensional and complementary morphological information of collagen fibers. Based on the multi-dimensional structural features, a support vector machine model is trained for the examination of breast cancer levels, which provides a more complete analysis of the structural information of breast tissue and improves the accuracy of breast cancer examination.

[0031] 4. Compared to traditional collagen fiber analysis tools that provide insufficient morphological information on collagen fibers, this invention proposes and implements a novel algorithm for extracting the cell-fiber relative angle in breast tissue images. This structural parameter is used to quantify the angle difference between collagen fibers and the central cell boundary in breast tissue. This structural feature varies significantly in breast tissues at different cancer levels and is an important reference for determining the level of breast cancer, providing effective information for the analysis of cancer levels in breast tissue.

[0032] 5. Compared with traditional histopathological methods, this invention refines the correlation between the multi-dimensional structural characteristics of collagen fibers and the level of breast tissue cancer, and performs cancer detection by establishing a mathematical model, which has higher accuracy, sensitivity and specificity, with a detection accuracy of up to 95% for breast tissue at different cancer levels.

[0033] 6. Compared to traditional collagen fiber analysis tools that cannot reflect local cancer characteristics in tissues, the fiber orientation, fiber direction variance, cell-fiber relative angle, and fiber density parameters used in this invention have pixel-level resolution, which can reflect local cancer characteristics in tissues and detect the generation and development of micro lesions. Attached Figure Description

[0034] Figure 1 A flowchart illustrating the construction of a breast cancer screening tool based on collagen fiber multivariate fusion analysis for this invention;

[0035] Figure 2 This is a schematic diagram of the optical imaging results and multi-dimensional structural feature extraction results of breast tissue according to the present invention: Figure 2 (a) is a multiphoton micrograph of collagen fibers in breast tissue; Figure 2 (b) is an image of the calculated fiber orientation information visualized using pseudo-color coding technology; Figure 2 (c)-(e) are images visualized using pseudo-color coding technology for fiber density, fiber orientation variance, and cell-fiber relative angle. The focus is on local features around the cell boundary, and the pseudo-color coding area is within 35 micrometers around the cell boundary. Figure 2 (f)-(h) are bar charts showing the results of this invention in quantifying fiber density, fiber orientation variance, and cell-fiber relative angle;

[0036] Figure 3 The following are the results of the analysis of three tumor-associated collagen signals (TACS1-3) in breast tissue according to this invention: Figure 3 (a) is a schematic diagram of the structures of three tumor-related collagen signals; Figure 3 (b) are multiphoton micrographs of three tumor-associated collagen signals; Figure 3 (c)-(e) are images visualized using pseudo-color coding technology, showing the fiber orientation variance, fiber density, and cell-fiber relative angle of three tumor-related collagen signals.

[0037] Figure 4 This is a schematic diagram illustrating the results of the present invention in determining the level of breast cancer: Figure 4 (a) Box plot comparing the multivariate parameters of fiber orientation variance, fiber density and cell-fiber relative angle of three tumor-associated collagen signals (TACS1-3) in breast tissue; Figure 4 (b) A three-dimensional scatter plot of different tumor-related collagen signals constructed using three parameters—fiber orientation variance, fiber density, and cell-fiber relative angle—as coordinate axes, along with the discrimination results of the original validation and cross-validation, with discrimination accuracies of 95% and 90%, respectively. Figure 4 (c) Detailed discrimination results of the breast cancer level discrimination model constructed using the above parameters. Detailed Implementation

[0038] To describe the present invention in more detail, the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] This invention relates to a breast cancer screening tool based on multivariate fusion analysis of collagen fibers. Its key feature is the fusion analysis of multivariate structural features of breast tissue to establish a breast cancer level discrimination model to assist in breast cancer screening. Figure 1 This is an overall flowchart of the present invention, which includes the following steps:

[0040] Optical imaging of collagen fibers in the extracellular matrix of breast tissue is mainly performed using two-photon fluorescence and second harmonic imaging techniques.

[0041] Multi-dimensional structural feature extraction was performed on collagen fibers to calculate quantitative results such as fiber orientation, fiber orientation variance, cell-fiber relative angle, and fiber density. Fiber orientation was obtained by calculating the difference between the spatial orientation of each collagen fiber pixel and its surrounding collagen fibers, used to quantify the pixel-level resolution of the collagen fiber arrangement direction in space, with a value range of 0–180°. Fiber orientation variance was obtained by calculating the variance of the spatial orientation of each collagen fiber pixel and its surrounding collagen fiber pixels, used to describe the degree of orderliness of the collagen fiber arrangement at the pixel-level resolution, with a value range of 0–1. A value closer to 0 indicates a more parallel and ordered arrangement of collagen fibers, while a value closer to 1 indicates a more scattered and disordered arrangement. The cell-fiber relative angle was calculated by... The difference in spatial orientation between a fibril pixel and its nearest neighbor cell boundary pixel is obtained. This difference quantifies the pixel-level resolution angular difference between the collagen fiber and the central cell boundary, ranging from 0 to 90°. A value of 0° indicates that the collagen fiber is completely parallel to the cell boundary, while a value of 90° indicates that the collagen fiber is perpendicular to the cell boundary. Fiber density is obtained by calculating the proportion of collagen fiber pixels within the neighborhood, quantifying the spatial distribution of collagen fiber pixel-level resolution. A value ranging from 0 to 1 indicates that the closer the value is to 0, the lower the collagen fiber content, and the closer the value is to 1, the higher the collagen fiber content. After obtaining the quantification results of the multi-dimensional structural features at the pixel-level resolution, pseudo-color coding technology is used to visualize the morphological information of collagen fibers provided by these parameters, thereby more intuitively and clearly displaying the structural features of collagen fibers.

[0042] Regions in breast tissue subjected to two-photon fluorescence / second harmonic imaging were marked, and the tissue was stained with hematoxylin and eosin for cancer pathological examination. The cancer level of the imaging region was marked according to the examination results.

[0043] Based on the calculated quantitative results of collagen fiber multi-dimensional structural features and cancer level marker information, a support vector machine model was established. This model calculates the cancer level of breast tissue using four collagen fiber multi-dimensional structural features: fiber orientation, fiber direction variance, cell-fiber relative angle, and fiber density. These multi-dimensional structural features provide multi-dimensional and complementary morphological information of collagen fibers, ensuring the accuracy of the model and enabling highly accurate analysis and examination of breast tissue cancer levels.

[0044] In the support vector machine model, each sample is defined as a vector based on the quantization result of its multivariate structural features. The quantification results of the multi-dimensional structural features include fiber orientation variance, cell-fiber relative angle, and fiber density, which are respectively derived from vi ,r i ,c i This indicates that the sample vector is composed of express.

[0045] Since breast cancer levels are classified into three categories, a three-class support vector machine (SVM) model is needed. However, the standard SVM model can only perform binary classification, so it needs to be extended. The approach to the three-class classification problem is to break it down into a series of binary classification problems, and then recombine these binary classifiers to obtain the three-class classification result. The steps for constructing the binary classification SVM model are as follows:

[0046] Based on the quantification results of multivariate structural features, the sample is defined as a vector. have Where v i ,r i ,c i Let V represent fiber orientation variance, cell-fiber relative angle, and fiber density, respectively. To distinguish between breast tissue samples of two different cancer levels, a segmentation hyperplane with the following form is sought from the sample space:

[0047]

[0048] in Let b be the normal vector of the segmenting hyperplane, and let b be the displacement of the segmenting hyperplane. The following discriminant is used to determine the cancer level of the breast tissue:

[0049]

[0050] Where y i It is the tag information, y i =1 indicates that the image belongs to type 1, y i =2 indicates that the image belongs to type 2. To obtain the values ​​of ω and b, a large amount of type-labeled sample data is needed for model training to find the optimal segmentation hyperplane. Specifically, this involves solving the following conditional extremum equation:

[0051]

[0052] Where st represents the condition that must be met simultaneously, and m is the total number of samples. Based on the obtained ω and b values, an optimal segmentation hyperplane is constructed to distinguish collagen fiber samples at two different cancer levels, thus establishing a binary classifier for breast cancer level.

[0053] It should be noted that the multivariate structural feature parameters include fiber orientation, fiber direction variance, cell-fiber relative angle, and fiber density. Among them, the value of fiber orientation is related to the selected reference frame, so this parameter was not used when building the support vector machine model. The calculation of the two parameters, fiber direction variance and cell-fiber relative angle, is based on fiber orientation, and these two parameters also contain the structural information provided by fiber orientation.

[0054] In addition, to provide more comprehensive morphological information on collagen fibers, a novel algorithm was proposed and implemented to extract the cell-fiber relative angle in breast tissue images. This structural parameter is used to quantify the angle difference between collagen fibers and the central cell boundary in breast tissue. This structural feature varies significantly in breast tissues at different cancer levels and is an important reference for determining the level of breast cancer, providing effective information for the analysis of cancer levels in breast tissue.

[0055] Through the above steps, pairwise training models are performed on the sample data of the three types of cancer to obtain three binary classifiers for each cancer level, whose segmentation hyperplane normal vectors are... The values ​​are: [0.37589792, 1.01374887, -0.34803396], [-1.33374423, -0.64543886, 2.45260154], [0.91955191, -0.5849421, -1.98512522], and the displacements b of the segmenting hyperplane are 0.30250453, 0.36193904, and 0.16197132, respectively. Each classifier is used to distinguish a certain cancer level type from other types.

[0056] For a breast tissue sample requiring breast cancer screening, optical imaging of the tissue's collagen fibers is first performed. Then, the multivariate structural feature parameters are calculated and input into the classification function of each binary classifier. The type of cancer level of the sample with the largest y-value is determined as the unknown collagen fiber sample, thus obtaining a breast cancer screening tool based on collagen fiber multivariate fusion analysis.

[0057] The specific embodiments of the present invention will be described below with reference to the accompanying drawings and specific implementation examples, but the application of the present invention is not limited to this.

[0058] This invention relates to a breast cancer screening tool based on collagen fiber multivariate fusion analysis. The specific steps are as follows:

[0059] (1) Optical imaging of breast tissue was performed to obtain images of collagen fibers in the extracellular matrix within the breast tissue. Optical imaging mainly used two-photon fluorescence and second harmonic imaging techniques. The procedure is as follows: Figure 1 As shown.

[0060] (2) Then, multi-dimensional structural features of the collagen fibers in the obtained breast tissue are extracted, and quantitative results such as fiber orientation, fiber direction variance, cell-fiber relative angle, and fiber density are calculated and obtained. The process is as follows: Figure 1 As shown. Fiber orientation is obtained by calculating the difference between the spatial orientation of each collagen fiber pixel and its surrounding collagen fibers. It is used to quantify the pixel-level resolution of collagen fiber arrangement in space, with a value ranging from 0 to 180°. Fiber orientation variance is obtained by calculating the variance of the spatial orientation of each collagen fiber pixel and its surrounding collagen fiber pixels. It is used to describe the degree of orderliness of the collagen fiber arrangement at the pixel-level resolution, with a value ranging from 0 to 1. The closer the value is to 0, the more parallel and ordered the collagen fibers are arranged; the closer the value is to 1, the more scattered and disordered the collagen fibers are arranged. The cell-fiber relative angle is calculated by... The difference in spatial orientation between a fibril pixel and its nearest neighbor cell boundary pixel is used to quantify the pixel-level resolution angular difference between the collagen fiber and the central cell boundary. The value ranges from 0 to 90°, where 0° indicates the collagen fiber is completely parallel to the cell boundary, and 90° indicates it is perpendicular. Fiber density is obtained by calculating the proportion of collagen fiber pixels within the neighborhood, used to quantify the spatial distribution of collagen fiber pixel-level resolution. The value ranges from 0 to 1, where a value closer to 0 indicates a lower collagen fiber content, and a value closer to 1 indicates a higher collagen fiber content. After obtaining the quantified results of the multi-dimensional structural features at pixel-level resolution, pseudo-color encoding technology is used to visualize the morphological information of collagen fibers provided by these parameters, thus more intuitively and clearly displaying the structural features of collagen fibers. A schematic diagram of the calculation and pseudo-color encoding results is shown below. Figure 2 As shown, Figure 2 (a) is a multiphoton micrograph of collagen fibers in breast tissue; Figure 2 (b) is an image of the calculated fiber orientation information visualized using pseudo-color coding technology; Figure 2 (c)-(e) are images visualized using pseudo-color coding technology for fiber density, fiber orientation variance, and cell-fiber relative angle. The focus is on local features around the cell boundary, and the pseudo-color coding area is within 35 micrometers around the cell boundary. Figure 2 (f)-(h) are bar charts showing the results of this invention's quantification of fiber density, fiber orientation variance, and cell-fiber relative angle.

[0061] (3) Hematoxylin-eosin staining of breast tissue is used for cancer pathological examination, and the cancer level of the multiphoton imaging area is marked according to the examination results. Three tumor-associated collagen signals (TACS1-3) in the breast can help identify the cancer level. TACS1 is defined by collagen fibers that are curved around the tumor and is a marker for locating small tumors. TACS2 is defined by stretched collagen fibers, which are arranged parallel to the tumor boundary due to tumor growth. TACS3 is defined by collagen fibers that are irregularly arranged perpendicular to the tumor boundary, and this state of collagen fibers promotes tumor migration. TACS is closely related to the remodeling of the collagen matrix and the occurrence and invasion of tumors. It is a potential biomarker for predicting the cancer level of breast tissue. Quantitative characterization and classification of TACS are of great significance for understanding the structure-function relationship and cancer level of breast tissue. Figure 3 This is a schematic diagram of the results of analyzing three tumor-associated collagen signals (TACS1-3) in breast tissue. Figure 3 (a) is a schematic diagram of the structures of three tumor-related collagen signals; Figure 3 (b) are multiphoton micrographs of three tumor-associated collagen signals; Figure 3 Images (c)-(e) are images visualized using pseudo-color encoding, showing the fiber orientation variance, fiber density, and cell-fiber relative angles of three tumor-related collagen signals. This invention enables the analysis of collagen fibers at pixel-level resolution. Combined with pseudo-color encoding technology, it can clearly and intuitively display the structural characteristics of collagen fibers in breast tissue at different cancer levels, helping users to grasp the structural differences between breast tissues at different cancer levels.

[0062] (4) Based on the calculated quantification results of collagen fiber multi-dimensional structural features and cancer level marker information, a sample vector is established according to the quantification results of multi-dimensional structural features. Where v i ,r i ,c i Let V represent fiber orientation variance, cell-fiber relative angle, and fiber density, respectively. To distinguish between two breast tissue samples with different cancer levels, the following discriminant is used to determine the cancer level of the breast tissue:

[0063]

[0064] Where y i It is the tag information, y i =1 indicates that the image belongs to type 1, y i =2 indicates that the image belongs to type 2.

[0065] Since breast cancer levels are classified into three categories, a three-class classification approach is adopted to establish a three-class support vector machine model. The model is trained pairwise on sample data of the three cancer levels to obtain three binary classifiers for each cancer level. The normal vector of the segmentation hyperplane is... The values ​​are: [0.37589792, 1.01374887, -0.34803396], [-1.33374423, -0.64543886, 2.45260154], [0.91955191, -0.5849421, -1.98512522], and the displacements b of the segmenting hyperplane are 0.30250453, 0.36193904, and 0.16197132, respectively. Each classifier is used to distinguish a certain cancer level type from other types. For a breast tissue that needs to be examined for breast cancer, optical imaging of the collagen fibers of the tissue is first performed, and then the results of its multivariate structural feature parameters are calculated and the values ​​are input into the classification function of each binary classifier. The category with the largest y-value is identified as the cancer level type of the unknown collagen fiber sample. When multiple samples' multivariate structural feature parameters are input, the model can output the cancer level discrimination results and scatter plots corresponding to different parameters. The scatter plots are used to more intuitively analyze the cancer development level of breast tissue samples. A schematic diagram and scatter plot of the results for breast cancer level discrimination are shown below. Figure 4 As shown. Figure 4 (a) A scatter plot comparing the multivariate parameters of fiber orientation variance, fiber density and cell-fiber relative angle in three different cancer levels of breast tissue samples is shown. It can be seen from the figure that the three structural parameters of samples at different cancer levels have obvious distinguishability, which shows the reliability of the present invention for breast cancer level examination. Figure 4 (b) A three-dimensional scatter plot of samples at three cancer levels is presented, constructed using three parameters—fiber orientation variance, fiber density, and cell-fiber relative angle—as coordinate axes. The plot shows that samples of the same cancer level are clustered, while samples of different cancer levels are independently dispersed, demonstrating the potential of this invention for examining different levels of breast cancer. The model's differentiation results for the three cancer levels are as follows: Figure 4 As shown in (b) and (c), the discrimination accuracy of the original validation and cross-validation is 95% and 90%, respectively, demonstrating the high accuracy, high specificity and sensitivity of the breast cancer screening tool proposed in this invention.

[0066] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A breast cancer examination method based on collagen fiber multi-element fusion analysis, characterized by The application discloses a breast cancer level discrimination model based on fusion analysis of multiple structural features of breast tissue, and provides a breast cancer examination method. 1) collagen fibers in extracellular matrix of breast tissue to be examined are imaged by two-photon fluorescence and second harmonic imaging technology, and micron resolution optical images are obtained; 2) multiple structural features of the collagen fiber optical images of the breast tissue are extracted, and quantitative results including fiber orientation, fiber direction variance, cell-fiber relative angle and fiber density are obtained; 3) hematoxylin-eosin staining is performed on the breast tissue for cancer pathology examination, and the cancer level of the two-photon fluorescence / second harmonic image obtained in step 1) is marked according to the examination result; 4) a support vector machine model is established based on the quantitative results of the multiple structural features calculated in step 2) and the image cancer level marking obtained in step 3); 5) In the SVM model, samples are defined as vectors according to their multivariate texture quantification results The multivariate texture quantification results include fiber direction variance, cell-fiber relative angle and fiber density, respectively represented by v i ,r i ,c i , and the vector in the model, a sample space is searched to separate breast tissue samples of two different cancer levels, and the equation of the separation hyperplane is as follows: wherein To segment the hyperplane, b is the displacement of the hyperplane, the following discriminant is used to determine the cancer level type to which the breast tissue belongs: where y i is the cancer level marker obtained in step 3), y i = 1 means that the image belongs to class 1, y i = 2 means that the image belongs to class 2; (6) the model is trained by using the image cancer level marking and the quantitative results of the multiple structural features, the best separation hyperplane is searched, and the values of ω and b are solved by solving the following equation: where s.t. represents that the conditions need to be met at the same time, and m is the total number of samples; 7) the best separation hyperplane is constructed according to the obtained values of ω and b, and is used to distinguish collagen fiber samples in two different cancer levels, and a breast cancer level binary classifier is established; 8) the cancer level type of an actual collagen fiber sample is divided into three categories, three breast cancer level binary classifiers are constructed based on the training method of steps 5) to 7), each classifier is used to distinguish a certain category from all other categories, when testing an unknown collagen fiber sample, the sample is input into each classification function, the category with the maximum function value is determined as the cancer level type of the unknown collagen fiber sample, and then a breast cancer examination method based on multiple fusion analysis of collagen fibers is obtained.

2. The breast cancer examination method based on collagen fiber multi-element fusion analysis according to claim 1, characterized in that, The fiber orientation in step 2) is obtained by calculating the difference between each collagen fiber pixel and the spatial orientation of the surrounding collagen fibers, and the resolution of the obtained fiber orientation quantitative result is pixel level; the fiber direction variance is obtained by calculating the variance between each collagen fiber pixel and the spatial orientation of the surrounding collagen fiber pixels, the resolution of the obtained fiber direction variance quantitative result is pixel level; the cell-fiber relative angle is obtained by calculating the difference between each collagen fiber pixel and the spatial orientation of the nearest cell boundary pixel, the resolution of the obtained cell-fiber relative angle quantitative result is pixel level; and the fiber density is obtained by calculating the proportion of collagen fiber pixels in the field, and the resolution of the obtained fiber density quantitative result is pixel level.

3. The breast cancer examination method based on collagen fiber multi-element fusion analysis according to claim 2, characterized in that, The fiber orientation is used for quantifying the arrangement direction of collagen fibers in space, with a value ranging from 0 to 180°; the fiber direction variance is used for describing the ordered degree of collagen fiber arrangement, with a value ranging from 0 to 1, and the closer to 0, the more parallel and ordered the collagen fiber arrangement, and the closer to 1, the more scattered and disordered the collagen fiber arrangement; the cell-fiber relative angle is used for quantifying the angle difference between the collagen fiber and the cell boundary, with a value ranging from 0 to 90°, and the value of 0° indicates that the collagen fiber is completely parallel to the cell boundary, and the value of 90° indicates that the collagen fiber is perpendicular to the cell boundary; the fiber density is used for quantifying the spatial distribution of collagen fibers, with a value ranging from 0 to 1, and the closer to 0, the less the collagen fiber content, and the closer to 1, the higher the collagen fiber content.

4. The collagen fiber multiplex fusion analysis-based breast cancer examination method according to claim 1 or 2 or 3, characterized by, The step 5) - 8) training obtained three breast cancer level binary classifier partition hyperplane normal vector [0.37589792, 1.01374887, -0.34803396], [-1.33374423, -0.64543886, 2.45260154], [0.91955191, -0.5849421, -1.98512522], the displacement b of the partition hyperplane is 0.30250453, 0.36193904, 0.16197132 respectively.

Images