Methods for identifying resection margins in breast cancer

Abstract

The present invention relates to a method for identifying breast cancer margins based on multiphoton tumor-related collagen features, comprising: obtaining a breast cancer multiphoton image; collecting a set number of image blocks in the breast cancer multiphoton image according to the tumor-related collagen features in the breast cancer multiphoton image, performing data cleaning and data preprocessing, and constructing a training data set; constructing a tissue component recognition module and using it to learn the feature information in the training data set; constructing a tissue component visualization module according to the pathological characteristics of breast tissue and the clinical tumor resection principle, and displaying the tissue component recognition results; performing local to overall recognition on a given breast cancer multiphoton image, inputting the image blocks into the tissue component recognition module in an overlapping manner in a sliding window manner, and outputting the tissue component categories; and finally integrating all tissue component recognition results to form a final margin recognition result. The method is conducive to quickly and automatically identifying the resection range and breast cancer margin position.

Description

Technical Field

[0001] The invention belongs to the field of computer vision, and in particular relates to a method for identifying a breast cancer resection margin. Background Art

[0002] Surgical resection is one of the main means of clinical treatment of breast cancer, and determining the extent of resection is of great significance for breast cancer resection. The existing method mainly relies on the experience of surgeons to make judgments, which has great uncertainty and often leads to contradictions between under-treatment and over-treatment. Although the effect of surgery can be verified by pathological examination after surgery, the traditional pathological examination method requires cumbersome steps such as sectioning and staining, and cannot provide surgeons with timely surgical guidance.

[0003] Multiphoton microscopy has been widely used in biomedical research. This technology can obtain high-resolution images for tissue pathology diagnosis without labeling, and has the potential to provide intraoperative guidance for surgeons. Studies have shown that multiphoton tumor-related collagen features are closely related to the development of breast cancer, providing an important scientific basis for accurately identifying breast cancer resection margins. However, direct identification of multiphoton images by human eyes requires not only a lot of training for clinicians, but also in-depth cooperation between optical experts and medical experts from multiple departments. Summary of the invention

[0004] The object of the present invention is to provide a method for identifying a breast cancer resection margin, which is conducive to quickly and automatically identifying the resection range and the breast cancer resection margin position.

[0005] To achieve the above object, the technical solution adopted by the present invention is: a method for identifying the resection margin of breast cancer, comprising the following steps:

[0006] Step S1: selecting an excitation wavelength according to the multiphoton excitation characteristics of breast tissue, and acquiring a multiphoton image of breast cancer corresponding to the excitation wavelength;

[0007] Step S2: according to the tumor-related collagen features in the breast cancer multiphoton image, a set number of image blocks are collected in the breast cancer multiphoton image, and data cleaning and data preprocessing are performed on the image blocks to construct a training data set;

[0008] Step S3: constructing a tissue component recognition module, and using the constructed tissue component recognition module to learn feature information in the training data set;

[0009] Step S4: constructing a tissue component visualization module based on the pathological characteristics of breast tissue and clinical tumor resection principles to display the tissue component identification results;

[0010] Step S5: The given breast cancer multiphoton image is identified from local to global perspectives, and image blocks are input into the tissue component identification module in an overlapping manner in a sliding window manner, and the tissue component category is output; finally, all tissue component identification results are integrated to form the final resection margin identification result.

[0011] Furthermore, in step S1, a multiphoton microscopic imaging system is used to obtain a multiphoton image of breast cancer. The multiphoton microscopic imaging system has a tunable output wavelength, obtains backscattered signals of tissue samples through a reflection grating and a photomultiplier tube array, and realizes large-scale scanning imaging through an electric workbench.

[0012] Furthermore, in step S1, unlabeled slices of breast cancer cases are collected and the tissue composition of the samples is analyzed; through multiple experimental comparisons, the excitation wavelength most suitable for breast tissue imaging is selected to produce images of the best quality; the tumor and its surrounding area in the slice are selected to collect multiphoton images of breast cancer.

[0013] Furthermore, step S2 specifically includes the following steps:

[0014] Step S21: based on the tumor-related collagen features and tumor cells, local sampling of important tissue components and auxiliary identification targets in the breast cancer multiphoton image is performed; the important tissue components include: tumor-related collagen features and tumor cells; the auxiliary identification targets include: fat, normal collagen, normal lobules, normal ducts, normal cells; a sufficient number of image blocks are collected by category to construct an initial data set;

[0015] Step S22: Perform data cleaning to remove image blocks with damaged samples and abnormal features in the initial data set;

[0016] Step S23: perform data preprocessing by randomly rotating, mirroring, and adjusting brightness of image blocks to expand the total amount of data and balance data distribution, thereby constructing a training data set.

[0017] Furthermore, step S3 specifically includes the following steps:

[0018] Step S31: constructing a tissue component recognition network with a residual structure, which is divided into three stages and consists of a downsampling layer, an intermediate layer and a classification layer; the input of the network is an image block X, and the output is a tissue component category Y;

[0019] The downsampling layer of the first stage consists of 1 convolution layer and 1 pooling layer. The original image block X is input into a convolution layer with a convolution kernel of 7×7 and an activation function of ReLU, and then passed through a pooling layer with a convolution kernel of 3×3 and an activation function of Max to complete the conversion from image to feature map and obtain the feature map F. The specific expression is:

[0020] F0=Max(Conv(X))

[0021] Where X represents the input image block;

[0022] The middle layer of the second stage consists of 8 series-connected residual modules, which are used to extract high-dimensional features that can describe breast tissue components. The residual module has the same structure, consisting of two convolutional layers with a convolution kernel of 3×3 and a step size of 1. The high-dimensional features of the image block are extracted by 8 series-connected residual modules. The calculation formula is:

[0023] S0=Conv(Conv(F0))+F0

[0024] S1=Conv(Conv(S0))+S0

[0025] …

[0026] Conv7=Conv(Conv(S6))+S6

[0027] Among them, S0, S1, ..., S7 represent the output features of each residual module, and the feature S7 output by the last residual module is used as the highest dimensional feature F1 extracted by the intermediate layer;

[0028] The classification layer of the third stage is composed of a pooling layer, a fully connected layer and a regression layer in series; the highest dimensional feature extracted by the middle layer is first input into a pooling layer with a convolution kernel of 1×1 and an activation function of Average to obtain an averaged high-dimensional feature; then through a fully connected layer with an activation function of ReLU, the discriminative high-dimensional features are integrated to obtain a one-dimensional vector; then through a regression layer with an activation function of Softmax for classification, finally the tissue component recognition result Y of the image block is obtained; the calculation formula is:

[0029] Y=Softmax(Average(Conv(F1)))

[0030] Among them, Y represents the output tissue component recognition result;

[0031] Step S32: According to the distribution of patients and data, the target categories in the training data set are divided into: training set, validation set and test set in a unified manner; the gradient of each parameter in the network is calculated using the back propagation method, and the parameters in the network are updated using the stochastic gradient descent method.

[0032] Furthermore, step S4 specifically includes the following steps:

[0033] Step S41: According to the pathological characteristics of breast tissue, the identification targets in the multiphoton image of breast cancer are divided into three sets: a tumor cell set, a tumor cell infiltration set, and an infiltration front set; wherein the tumor cell set includes: tumor-related collagen features 1-3 and tumor cells; the tumor cell infiltration set includes: tumor-related collagen features 4-6; the infiltration front set includes: tumor-related collagen features 7-8, fat, normal collagen, normal lobules, normal ducts, and normal cells;

[0034] Step S42: According to the clinical tumor resection principle, the three types of sets are divided into resection areas and retention areas, and a recognition result visualization module is designed to distinguish them; the tumor cell set and the tumor cell infiltration set are set as the resection areas, and warm colors are used to represent different recognition targets; the infiltration front set is set as the retention area, and cold colors are used to represent different recognition targets.

[0035] Furthermore, step S5 specifically includes the following steps:

[0036] Step S51: In a given breast cancer multiphoton image, each image block of the breast cancer multiphoton image is input into the trained tissue component recognition module in an overlapping manner in a sliding window manner, and the tissue component Y is output. i , and the number of recognition times N of the area covered by the image block is increased by 1;

[0037] Step S52: According to tissue component Y i Assign a value to the area covered by the image block, the cut-off area is 1, and the retained area is 0, and the assignment result T is obtained. i ;

[0038] Step S53: Integrate the assignment results, and take the average of the assignment results of each region according to the number of overlapping identifications N of each region to obtain the final cutting edge identification result Z. The calculation formula is:

[0039] Z=∑T i / N

[0040] Among them, the value of Z is between [0,1];

[0041] According to the cutting edge recognition result Z of each area, a cutting edge visualization module is designed, and three different colors are set to display the cutting edge, the resection area, and the retained area respectively; when Z is 0.5, it indicates that the area is the junction between the resection area and the retained area, that is, the cutting edge, which is displayed in yellow; when the value of Z is between (0.5, 1], it indicates that the area is the resection area, which is displayed in red; when the value of Z is between [0, 0.5), it indicates that the area is the retained area, which is displayed in blue.

[0042] Compared with the prior art, the present invention has the following beneficial effects: it provides a method for identifying breast cancer margins based on multiphoton tumor-associated collagen features. The method utilizes multiphoton tumor-associated collagen features to diagnose breast tissue sections, which solves the limitation of previous methods that require a lot of manpower and time costs, and realizes automatic identification of breast tissue components and breast cancer margin positions in multiphoton images, effectively improving the efficiency of breast cancer margin identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of a method implementation of an embodiment of the present invention;

[0044] Figure 2 is a multiphoton image of breast cancer in an embodiment of the present invention;

[0045] Figure 3 is the breast tissue component identification result in the embodiment of the present invention;

[0046] Figure 4 This is the breast cancer resection margin identification result in the embodiment of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0049] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0050] like Figure 1 As shown, this embodiment provides a method for identifying breast cancer resection margin based on multiphoton tumor-related collagen features, comprising the following steps:

[0051] Step S1: Select a suitable excitation wavelength according to the multiphoton excitation characteristics of breast tissue, and obtain a multiphoton image of breast cancer corresponding to the excitation wavelength.

[0052] Step S2: According to the tumor-related collagen features in the breast cancer multiphoton images, sufficient image blocks are collected in the multiphoton images, and data cleaning and data preprocessing are performed to construct a training data set.

[0053] Step S3: Design a tissue component recognition module, and use the designed recognition module to learn the feature information in the training data set.

[0054] Step S4: According to the pathological characteristics of breast tissue and clinical tumor resection principles, a tissue component visualization module is designed to display the tissue component identification results.

[0055] Step S5: Perform local to global recognition on the given breast cancer multiphoton image, input the image blocks into the tissue component recognition module in an overlapping manner in a sliding window manner, and output the tissue component category. Finally, integrate all tissue component recognition results to form the final resection margin recognition result.

[0056] In this embodiment, a label-free multiphoton microscopic imaging system is used in step S1 to obtain a multiphoton image of breast cancer; the multiphoton microscopic imaging system is built on a commercial laser scanning microscope platform (LSM 880 Zeiss, Germany), uses a mode-locked femtosecond Ti:Sapphire laser as a light source, obtains backscattered signals of tissue samples through a reflection grating and a photomultiplier tube array, and realizes large-scale scanning imaging through an electric workbench. Unlabeled slices of breast cancer cases are collected, the tissue composition of the samples is analyzed, and the output wavelength of the laser is tuned according to the imaging effect; through multiple experimental comparisons, 810nm is selected as the most suitable excitation wavelength for breast tissue imaging to produce the best image quality; Figure 2 As shown, the tumor and its surrounding area in the slice are selected, and multiphoton images of breast cancer with a size of 4000×4000 pixels to 6000×6000 pixels are collected.

[0057] In this embodiment, step S2 specifically includes the following steps:

[0058] Step S21: Based on the tumor-related collagen features and tumor cells, local sampling of important tissue components and auxiliary identification targets in breast cancer multiphoton images is performed. Each local image block has the same size of 512×512 pixels. Among them, important tissue components include: tumor-related collagen features 1-8 and tumor cells; auxiliary identification targets include: fat, normal collagen, normal lobules, normal ducts, normal cells, etc. About 1,000 local image blocks are collected by category to construct the initial data set.

[0059] Step S22: Perform data cleaning to remove image blocks with damaged samples and abnormal features in the initial data set.

[0060] Step S23: perform data preprocessing, use functions such as cv2.getRotationMatrix2D and cv2.flip to perform random rotation, mirroring, brightness adjustment and other operations on the image blocks, expand each type of data to about 2000 images, and construct a training data set.

[0061] In this embodiment, step S3 specifically includes the following steps:

[0062] Step S31: Design a tissue component recognition network with a residual structure, which is divided into three stages and consists of a downsampling layer, an intermediate layer and a classification layer. The input of the network is an image block X, and the output is a tissue component category Y.

[0063] The first stage of downsampling layer consists of 1 convolution layer and 1 pooling layer. The original image block X is input into a convolution layer with a convolution kernel of 7×7 and an activation function of ReLU, and then through a pooling layer with a convolution kernel of 3×3 and an activation function of Max to complete the conversion from image to feature map and obtain the feature map F. The specific expression is:

[0064] F0=Max(Conv(X))

[0065] Where X represents the input image patch.

[0066] The middle layer of the second stage consists of 8 series-connected residual modules, which are used to extract high-dimensional features that can describe breast tissue components. The residual module has the same structure, consisting of two convolutional layers with a convolution kernel of 3×3 and a step size of 1. The high-dimensional features of the image block are extracted by 8 series-connected residual modules, and the calculation formula is:

[0067] S0=Conv(Conv(F0))+F0

[0068] S1=Conv(Conv(S0))+S0

[0069] …

[0070] S7=Conv(Conv(S6))+S6

[0071] Among them, S0, S1, ..., S7 represent the output features of each residual module, and the feature S7 output by the last residual module is used as the highest dimensional feature F1 extracted by the intermediate layer.

[0072] The classification layer of the third stage is composed of a pooling layer, a fully connected layer and a regression layer in series. First, the highest dimensional feature extracted by the middle layer is input into a pooling layer with a convolution kernel of 1×1 and an activation function of Average to obtain an averaged high-dimensional feature; then, through a fully connected layer with an activation function of ReLU, the high-dimensional features with discriminative features are integrated to obtain a one-dimensional vector; then, through a regression layer with an activation function of Softmax, classification is performed, and finally the tissue component recognition result Y of the image block is obtained. The calculation formula is:

[0073] Y=Softmax(Average(Conv(F1)))

[0074] Among them, Y represents the output tissue component recognition result;

[0075] Step S32: According to the distribution of patients and data, the target categories in the training data set are divided into training set, validation set and test set according to the patient number and the ratio of 6:2:2. The gradient of each parameter in the classification network is calculated by the back propagation method, and the parameters in the classification network are updated by the stochastic gradient descent method.

[0076] In this embodiment, step S4 specifically includes the following steps:

[0077] Step S41: According to the pathological characteristics of breast tissue, the identification targets in the breast cancer multiphoton image are divided into three sets: tumor cell set, tumor cell infiltration set, and infiltration front set. Among them, the tumor cell set includes: tumor-related collagen features 1-3 and tumor cells. The tumor cell infiltration set includes: tumor-related collagen features 4-6. The infiltration front set includes: tumor-related collagen features 7-8, fat, normal collagen, normal lobules, normal ducts and normal cells;

[0078] Step S42: According to the clinical tumor resection principle, the three types of sets are divided into resection areas and retention areas, and a recognition result visualization module is designed to distinguish them. Figure 3 As shown, the tumor cell set and the tumor cell infiltration set are set as the resection area, and warm colors are used to represent different recognition targets; the infiltration front set is set as the retention area, and cold colors are used to represent different recognition targets.

[0079] In this embodiment, step S5 specifically includes the following steps:

[0080] Step S51: For a given breast cancer multiphoton image, the breast cancer multiphoton image is divided into image blocks of 512×512 pixels in an overlapping manner in a sliding window manner. The image blocks are input into the trained tissue component recognition model, and the tissue component Y is output. i , and the recognition times N of the area covered by the image block is increased by 1. Among them, the overlap rate of the sliding window is 25%, and the maximum recognition times of the area covered by the image block is 16.

[0081] Step S52: According to tissue component Y i Assign a value to the area covered by the image block, the cut-off area is 1, and the retained area is 0, and the assignment result T is obtained. i ;

[0082] Step S53: Integrate the assignment results, and take the average of the assignment results of each region according to the number of overlapping identifications N of each region to obtain the final cutting edge identification result Z. The calculation formula is:

[0083] Z=∑T i / N

[0084] Among them, the value of Z is between p0,1].

[0085] Set the dot matrix M to be the set of all the cutting edge identification results Z. Through the cv2.resize function, M is enlarged by bicubic interpolation to obtain a dot matrix N that is sufficient to cover the given breast cancer multiphoton image. There is a transition trend between adjacent elements of the dot matrix N, and a smooth cutting edge curve can be obtained. According to the dot matrix N, a cutting edge visualization module is designed, and three different colors are set to display the cutting edge, the resection area, and the reserved area respectively. Figure 4 As shown in the figure, when Z is 0.5, it means that the area is the boundary between the resection area and the retained area, that is, the cutting edge, which is displayed in yellow; when the value of Z is between [0,0.5], it means that the area is the retained area, which is displayed in blue; when the value of Z is between (0.5,1], it means that the area is the resection area, which is displayed in red.

[0086] The above program design scheme provided in this embodiment can be stored in a computer-readable storage medium in a coded form and implemented in the form of a computer program, and the basic parameter information required for calculation is input through computer hardware, and the calculation result is output.

[0087] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] The present invention is described with reference to flowcharts of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process in the flowchart, as well as the combination of processes in the flowchart, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A device that specifies functions in a process or multiple processes.

[0089] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple functions specified in a flowchart.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

[0091] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for identifying breast cancer resection margin, characterized in that: The following steps are involved: Step S1: selecting an excitation wavelength according to the multiphoton excitation characteristics of breast tissue, and acquiring a multiphoton image of breast cancer corresponding to the excitation wavelength; Step S2: according to the tumor-related collagen features in the breast cancer multiphoton image, a set number of image blocks are collected in the breast cancer multiphoton image, and data cleaning and data preprocessing are performed on the image blocks to construct a training data set; Step S3: constructing a tissue component recognition module, and using the constructed tissue component recognition module to learn feature information in the training data set; Step S4: constructing a tissue component visualization module based on the pathological characteristics of breast tissue and clinical tumor resection principles to display the tissue component identification results; Step S5: Recognize the given breast cancer multiphoton image from local to overall, input the image blocks into the tissue component recognition module in an overlapping manner in a sliding window manner, and output the tissue component category; finally, integrate all tissue component recognition results to form the final resection margin recognition result; Step S3 specifically includes the following steps: Step S31: constructing a tissue component recognition network with a residual structure, which is divided into three stages and consists of a downsampling layer, an intermediate layer and a classification layer; the input of the network is an image block X, and the output is a tissue component category Y; The downsampling layer of the first stage consists of 1 convolution layer and 1 pooling layer. The original image block X is input into a convolution layer with a convolution kernel of 7×7 and an activation function of ReLU, and then passed through a pooling layer with a convolution kernel of 3×3 and an activation function of Max to complete the conversion from image to feature map and obtain the feature map F0. The specific expression is: F0=Max(Conv(X)) Where X represents the input image block; The middle layer of the second stage consists of 8 series-connected residual modules, which are used to extract high-dimensional features that can describe breast tissue components. The residual module has the same structure, consisting of two convolutional layers with a convolution kernel of 3×3 and a step size of 1. The high-dimensional features of the image block are extracted by 8 series-connected residual modules. The calculation formula is: S0=Conv(Conv(F0)+F0 S1=Conv(Conv(S0))+S0 …… S7=Conv(Conv(S6))+S6 Among them, S0, S1, ..., S7 represent the output features of each residual module, and the feature S7 output by the last residual module is used as the highest dimensional feature F1 extracted by the intermediate layer; The classification layer of the third stage is composed of a pooling layer, a fully connected layer and a regression layer in series; the highest dimensional feature extracted by the middle layer is first input into a pooling layer with a convolution kernel of 1×1 and an activation function of Average to obtain an averaged high-dimensional feature; then through a fully connected layer with an activation function of ReLU, the discriminative high-dimensional features are integrated to obtain a one-dimensional vector; then through a regression layer with an activation function of Softmax for classification, finally the tissue component recognition result Y of the image block is obtained; the calculation formula is: Y=Softmax(Average(Conv(F1))) Among them, Y represents the output tissue component recognition result; Step S32: Divide the target categories in the training data set into: training set, validation set and test set in a unified manner; calculate the gradient of each parameter in the network using the back propagation method, and update the parameters in the network using the stochastic gradient descent method; Step S4 specifically includes the following steps: Step S41: According to the pathological characteristics of breast tissue, the identification targets in the multiphoton image of breast cancer are divided into three sets: a tumor cell set, a tumor cell infiltration set, and an infiltration front set; wherein the tumor cell set includes: tumor-related collagen features 1-3 and tumor cells; the tumor cell infiltration set includes: tumor-related collagen features 4-6; the infiltration front set includes: tumor-related collagen features 7-8, fat, normal collagen, normal lobules, normal ducts, and normal cells; Step S42: According to the clinical tumor resection principle, the three types of sets are divided into resection areas and retention areas, and a recognition result visualization module is designed to distinguish them; the tumor cell set and the tumor cell infiltration set are set as the resection areas, and warm colors are used to represent different recognition targets; the infiltration front set is set as the retention area, and cold colors are used to represent different recognition targets.

2. The breast cancer resection margin identification method according to claim 1, characterized in that: In step S1, a multiphoton microscopic imaging system is used to obtain a multiphoton image of breast cancer. The multiphoton microscopic imaging system has a tunable output wavelength, obtains backscattered signals of tissue samples through a reflection grating and a photomultiplier tube array, and realizes large-scale scanning imaging through an electric workbench.

3. The breast cancer resection margin identification method according to claim 1, characterized in that: In step S1, unlabeled slices of breast cancer cases are collected and the tissue composition of the samples is analyzed; through multiple experimental comparisons, the excitation wavelength most suitable for breast tissue imaging is selected to produce images of the best quality; the tumor and its surrounding area in the slice are selected to collect multiphoton images of breast cancer.

4. The method for identifying breast cancer resection margin according to claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21: based on the tumor-related collagen features and tumor cells, local sampling of important tissue components and auxiliary identification targets in the breast cancer multiphoton image is performed; the important tissue components include: tumor-related collagen features and tumor cells; the auxiliary identification targets include: fat, normal collagen, normal lobules, normal ducts, normal cells; a sufficient number of image blocks are collected by category to construct an initial data set; Step S22: Perform data cleaning to remove image blocks with damaged samples and abnormal features in the initial data set; Step S23: perform data preprocessing by randomly rotating, mirroring, and adjusting brightness of image blocks to expand the total amount of data and balance data distribution, thereby constructing a training data set.

5. The method for identifying breast cancer resection margin according to claim 1, characterized in that: Step S5 specifically includes the following steps: Step S51: In a given breast cancer multiphoton image, each image block of the breast cancer multiphoton image is input into the trained tissue component recognition module in an overlapping manner in a sliding window manner, and the tissue component Y is output. i , and the number of recognition times N of the area covered by the image block is increased by 1; Step S52: According to tissue component Y i Assign a value to the area covered by the image block, the cut-off area is 1, and the retained area is 0, and the assignment result T is obtained. i ; Step S53: Integrate the assignment results, and take the average of the assignment results of each region according to the number of overlapping identifications N of each region to obtain the final cutting edge identification result Z. The calculation formula is: Z=∑T i / N Among them, the value of Z is between [0,1]; According to the cutting edge recognition result Z of each area, a cutting edge visualization module is designed, and three different colors are set to display the cutting edge, the resection area, and the retained area respectively; when Z is 0.5, it indicates that the area is the junction between the resection area and the retained area, that is, the cutting edge, which is displayed in yellow; when the value of Z is between (0.5, 1], it indicates that the area is the resection area, which is displayed in red; when the value of Z is between [0, 0.5), it indicates that the area is the retained area, which is displayed in blue.

Images