A tumor resection margin design method and system based on local skin image

Abstract

The invention relates to the technical field of medical image processing, and discloses a tumor margin design method and system based on a local skin image. The method comprises the following steps: performing random non-overlapping sampling on an initial data set to obtain a sampled data set; obtaining a human body part label; selecting a local small image, and using the coordinate information of the local small image to solve a reference coordinate range of a skin tumor margin image to be designed; obtaining a rough skin grain direction of the skin tumor margin image to be designed based on the reference coordinate range and the human body part label; obtaining a super-resolution magnified image from a super-resolution magnification model, and performing skin grain detection on the super-resolution magnified image to obtain an accurate skin grain line; performing tumor boundary detection on the skin tumor margin image to be designed to obtain tumor boundary information, and performing margin design on a skin tumor in the skin tumor margin image to be designed based on the tumor boundary information and the accurate skin grain line. The invention solves the problem of low accuracy in the existing tumor margin design.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a tumor resection margin design method and system based on local skin images. Background Art

[0002] Designing the incision margin for skin tumors is an important clinical issue. The design direction of the incision margin should be consistent with the direction of the patient's skin lines. Since the skin elasticity of patients of different ages and genders is different, the size of the incision should also take into account the patient's age and gender. Reasonable incision margin design can reduce skin surgical trauma, accelerate postoperative healing, and avoid scar formation.

[0003] The traditional method is for dermatologists to design the cutting edge directly on the patient's skin by referring to the human dermatoglyphics map and combining factors such as the patient's age and gender. This method relies heavily on the doctor's experience and is highly subjective. Therefore, the design of some cutting edges may lack rationality, which is not conducive to the patient's postoperative recovery. Tumor cutting edge design based on skin images provides qualitative and quantitative solutions. However, the skin tumor images taken are only partial photos, and the human eye cannot accurately locate the specific location and dermatoglyphic direction. It can be seen that the existing tumor cutting edge design has the problem of low accuracy. Summary of the invention

[0004] The present invention provides a tumor resection margin design method and system based on local skin images, so as to solve the problem of low accuracy in existing tumor resection margin design.

[0005] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0006] In a first aspect, the present invention provides a tumor resection margin design method based on a local skin image, comprising:

[0007] Acquire clinical skin images of different parts of the human body and accurate part labels corresponding to the clinical skin images to construct an initial data set, and perform random non-overlapping sampling on the initial data set to obtain a sampled data set;

[0008] Constructing a super-resolution magnification model and a human body part recognition model based on the sampled data set;

[0009] Acquire a small local image of the skin tumor resection margin to be designed, and input the small local image of the skin tumor resection margin to be designed into the human body part recognition model to obtain a human body part label;

[0010] Based on the human body part label, a similarity comparison is performed between the local small image of the skin tumor resection margin to be designed and the local small images in the sampling data set, and the reference coordinate range of the local small image of the skin tumor resection margin to be designed is solved by using the coordinate information of the local small image with high similarity in the sampling data set;

[0011] Obtaining a rough skin grain line of a local small image of a skin tumor resection margin to be designed based on the human body part label and the reference coordinate range;

[0012] Inputting a small local image of the skin tumor resection margin to be designed into the super-resolution magnification model to obtain a super-resolution magnified image, and performing skin texture detection on the super-resolution magnified image to obtain accurate skin grain lines;

[0013] Tumor boundary detection is performed on the local small image of the skin tumor resection margin to be designed to obtain tumor boundary information, and the skin tumor resection margin is designed in the local small image of the skin tumor resection margin to be designed based on the tumor boundary information and the precise skin grain line.

[0014] Optionally, constructing a super-resolution magnification model and a human body part recognition model based on the sampled data set includes:

[0015] A super-resolution magnification model and a human body part recognition model based on a convolutional neural network are constructed, and the super-resolution magnification model and the human body part recognition model are trained respectively using the sampled data set to obtain a trained super-resolution magnification model and a trained human body part recognition model.

[0016] Optionally, the skin clinical images of different parts of the human body and the accurate part labels corresponding to the skin clinical images are obtained to construct an initial data set. ,include:

[0017] Obtain skin clinical images of different parts of the human body and the accurate part labels corresponding to the skin clinical images to build the initial data set , where accurate part labels include: face, head, neck, chest, and back;

[0018] The initial dataset is:

[0019] ;

[0020] in, and They represent the skin clinical image and the part label corresponding to the skin clinical image, is the sequence number of the image, is the number of images.

[0021] Optionally, the initial data set Clinical images of the skin in Perform random non-overlapping sampling to obtain the sampled data set ,include:

[0022] Clinical images in skin Random Intercept Small local images , and clinical images of the skin The upper left corner is the origin of the coordinate system;

[0023] Record all local small images The coordinates of the upper left corner in the coordinate system, and the coordinates are calculated to make Small local images Overlay skin clinical images All pixel information in the image, the coordinate calculation is as follows:

[0024] , ∈Z and ;

[0025] in, represents the number of local small images, H represents the side length of the skin clinical image, h represents the side length of the local small image, Indicates rounding up;

[0026] Construct a sampling data set based on the local small image :

[0027] ;

[0028] in, The coordinate information of each local small image is recorded in.

[0029] Optionally, the method of using the coordinate information of the local small images with high similarity in the sampled data set to solve the reference coordinate range of the local small image of the skin tumor resection margin to be designed includes:

[0030] Calculate the local small image of the skin tumor resection margin to be designed With the sampled data set Zhongyu Small local images with the same human body part labels structural similarity, Pearson correlation coefficient and Manhattan distance;

[0031] The weighted sum of the calculation results is used to obtain a small local image of the skin tumor margin to be designed. With local small image The similarity value between , and its calculation formula is as follows:

[0032] ;

[0033] in, Indicates structural similarity, represents the Pearson correlation coefficient, represents the Manhattan distance, , , are fitting parameters.

[0034] Optionally, the method uses a small local image of the skin tumor resection margin to be designed With local small image The similarity information between them is used to solve the local small image of the skin tumor resection margin to be designed The reference coordinate range includes:

[0035] Extracting the dataset Small local image of the cutaneous tumor margin to be designed Similarity value between The highest front Small local images The coordinate information of the skin tumor is obtained by solving the mean value and standard deviation of the coordinate information, and the local small image of the skin tumor resection margin to be designed is obtained based on the mean value and standard deviation of the coordinate information. Reference coordinate range: ;

[0036] in, Represents the average value of coordinate information, Indicates the standard deviation of the coordinate information.

[0037] Optionally, obtaining a rough skin grain line of a local small image of a skin tumor resection margin to be designed based on the human body part label and the reference coordinate range includes:

[0038] Using a small local image of the skin tumor margin to be designed Reference coordinate range , compare the human dermatoglyphics atlas of the corresponding human body parts, and obtain the local small image of the skin tumor resection margin to be designed The rough direction of the skin grain.

[0039] Optionally, performing skin texture detection on the super-resolution magnified image to obtain accurate skin grain lines includes:

[0040] A random seed point is set on the rough skin grain line in the super-resolution magnified image, and the adjacent pixel points within 360° of the seed point are used as the search range. The similarity between the random seed point and the adjacent pixel points in the search range is calculated as shown in the following formula:

[0041]

[0042] in, , as well as and , as well as are the three color channel values ​​of the searched pixel and seed point in the CIELab color space respectively;

[0043] Taking any two adjacent pixels in the search range as comparison units, calculate the similarity values ​​between the adjacent pixels in the comparison unit and the random seed points, and sum them up as the similarity value of the comparison unit;

[0044] Calculate the distances between two adjacent pixels in the comparison unit and other pixels on the rough dermatoglyphic line within the search range that are closest to the comparison unit pixel, excluding the seed pixel, and sum them up as the distance measurement value of the comparison unit;

[0045] The self-correction metric value of the comparison unit is obtained by adding the similarity value of the comparison unit to the inverse of the distance metric value, the self-correction metric values ​​of all comparison units in the search range are calculated, and the sizes are compared, and the seed point is updated to the comparison unit with the largest self-correction metric value;

[0046] Repeat the above steps until the random seed points are on both sides of the rough dermatoglyphics, and use the movement trajectory of the random seed points in the self-correction process as the precise dermatoglyphics line.

[0047] Optionally, the step of designing a resection margin of a skin tumor in a small local image for designing a resection margin of the skin tumor based on the tumor boundary information and the precise skin grain line includes:

[0048] The tumor center was used as the cutting edge center, and the cutting edge was designed according to the precise skin grain line direction. The cutting edge design was corrected again based on the patient's gender.

[0049] If the patient is male and is , then the secondary correction margin width is:

[0050] ;

[0051] In the formula, Indicates the width of the secondary correction margin in male patients. Indicates the age of male patients, Indicates the width of the cutting edge;

[0052] If the patient is female and is , then the secondary correction margin width is:

[0053] ;

[0054] Indicates the width of the secondary correction margin in female patients. Indicates the age of female patients, Indicates the width of the cutting edge.

[0055] In a second aspect, an embodiment of the present application provides a tumor margin design system based on a local skin image, including a processor and a memory;

[0056] Memory, used to store computer programs;

[0057] The processor is used to implement any method step described in the first aspect when executing a program stored in the memory.

[0058] Beneficial effects:

[0059] The tumor margin design method based on local skin images provided by the present invention constructs an initial data set by acquiring clinical skin images of different parts of the human body and accurate part labels corresponding to the clinical skin images, and performs random non-overlapping sampling on the initial data set to obtain a sampled data set; the skin tumor margin image to be designed is input into a trained human body part recognition model to obtain the human body part label corresponding to the skin tumor margin image to be designed; a local small image similar to the skin tumor margin image to be designed in the sampled data set is selected through similarity comparison, and the coordinate information of the local small image is used to solve the reference coordinate range of the skin tumor margin image to be designed; based on the reference coordinate range and the human body part label, the rough skin texture of the skin tumor margin image to be designed is obtained. Direction: Input the image of the skin tumor margin to be designed into the trained super-resolution magnification model to obtain a super-resolution magnified image, and perform skin texture detection on the super-resolution magnified image to obtain precise skin grain lines; perform tumor boundary detection on the image of the skin tumor margin to be designed to obtain tumor boundary information, and perform skin margin design on the skin tumor in the image of the skin tumor margin to be designed based on the tumor boundary information and precise skin grain lines; directly predict the reference coordinate range of the source site and the location from the local skin image, and first obtain the rough skin grain lines by comparing with the human skin grain, and then obtain the precise skin grain lines by correcting the skin grain lines in the super-resolution image, and the incision size can be quantitatively designed in combination with the patient's age and gender, so as to provide accurate quantitative and qualitative reference information for skin tumor margin design. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart of a method for designing a tumor resection margin based on a local skin image according to a preferred embodiment of the present invention;

[0061] Figure 2 A schematic diagram of a random sampling method according to a preferred embodiment of the present invention;

[0062] Figure 3 Schematic diagram of a method for obtaining rough dermatoglyphic lines according to a preferred embodiment of the present invention, wherein (a) is a schematic diagram of a human dermatoglyphic map of the face and neck, and (b) is a schematic diagram of a rough dermatoglyphic direction of a local small image obtained by comparing the human dermatoglyphic map;

[0063] Figure 4 Schematic diagram of self-correction of a preferred embodiment of the present invention, wherein (a) is a schematic diagram of a rough dermatoglyphic direction in a super-resolution image, and (b) is a schematic diagram of a precise dermatoglyphic direction in a super-resolution image;

[0064] Figure 5 A schematic diagram of a method for obtaining accurate dermatoglyphics according to a preferred embodiment of the present invention;

[0065] Figure 6 Schematic diagram of tumor resection margin design according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0066] The technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not indicate quantity restrictions, but indicate the existence of at least one. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.

[0068] See also Figure 1-6 The present application embodiment provides a tumor resection margin design method based on a local skin image, comprising:

[0069] Acquire clinical skin images of different parts of the human body and accurate part labels corresponding to the clinical skin images to construct an initial data set, and perform random non-overlapping sampling on the initial data set to obtain a sampled data set;

[0070] Constructing a super-resolution magnification model and a human body part recognition model based on the sampled data set;

[0071] Acquire a small local image of the skin tumor resection margin to be designed, and input the small local image of the skin tumor resection margin to be designed into the human body part recognition model to obtain a human body part label;

[0072] Based on the human body part label, a similarity comparison is performed between the local small image of the skin tumor resection margin to be designed and the local small images in the sampling data set, and the reference coordinate range of the local small image of the skin tumor resection margin to be designed is solved by using the coordinate information of the local small image with high similarity in the sampling data set;

[0073] Obtaining a rough skin grain line of a local small image of a skin tumor resection margin to be designed based on the human body part label and the reference coordinate range;

[0074] Inputting a small local image of the skin tumor resection margin to be designed into the super-resolution magnification model to obtain a super-resolution magnified image, and performing skin texture detection on the super-resolution magnified image to obtain accurate skin grain lines;

[0075] Tumor boundary detection is performed on the local small image of the skin tumor resection margin to be designed to obtain tumor boundary information, and the skin tumor resection margin is designed in the local small image of the skin tumor resection margin to be designed based on the tumor boundary information and the precise skin grain line.

[0076] In the above embodiment, an initial data set is constructed by acquiring clinical skin images of different parts of the human body and accurate part labels corresponding to the clinical skin images, and random non-overlapping sampling is performed on the clinical skin images in the initial data set to obtain a sampled data set; a small local image of the resection edge of the skin tumor to be designed is input into the trained human body part recognition model to obtain the human body part label corresponding to the small local image of the skin tumor to be designed.

[0077] At the same time, by comparing the similarity between the local small image of the skin tumor cutting edge to be designed and the local small image with the same human body part label in the sampling data set, and using the coordinate information of the local small image with high similarity in the sampling data set to solve the reference coordinate range of the local small image of the skin tumor cutting edge to be designed; based on the human body part label and the reference coordinate range, the rough skin grain line of the local small image of the skin tumor cutting edge to be designed is obtained; the local small image of the skin tumor cutting edge to be designed is input into the trained super-resolution enlargement model to obtain a super-resolution enlarged image, and skin texture detection is performed on the super-resolution enlarged image, so as to realize the detection of the rough skin grain line. The system can perform self-correction to obtain precise dermatoglyphic lines; perform tumor boundary detection on the local small image of the skin tumor resection margin to obtain tumor boundary information, and design the resection margin of the skin tumor in the local small image of the skin tumor resection margin based on the tumor boundary information and precise dermatoglyphic lines; directly predict the reference coordinate range of the source site and the location from the local skin image, and first obtain the rough dermatoglyphic lines by comparing with the human dermatoglyphics, and then obtain the precise dermatoglyphic lines by correcting the dermatoglyphics in the super-resolution image, and can quantitatively design the incision size based on the patient's age and gender, providing accurate quantitative and qualitative reference information for the design of skin tumor resection margin.

[0078] In the embodiment, the local small image and the accurate part label are both synonyms for the acquired data. The small image in the local small image refers to a relatively small image compared to the skin clinical image, so it is called a local small image. The accurate part label means that the acquisition of the label needs to correspond to the accurate part.

[0079] Optionally, the step of acquiring clinical skin images of different parts of a human body and accurate part labels corresponding to the clinical skin images to construct an initial data set includes:

[0080] Obtain skin clinical images of different parts of the human body and accurate part labels corresponding to the skin clinical images to construct an initial data set, where the part labels include: face, head, neck, chest, back and other parts of the human body;

[0081] The initial dataset is:

[0082] ;

[0083] in, and They represent the skin clinical image and the part label corresponding to the skin clinical image, is the sequence number of the image, is the number of images.

[0084] Optionally, for the initial dataset Clinical images of the skin in Random non-overlapping sampling is performed to obtain a sampled data set, including:

[0085] Clinical images in skin Random Intercept Small local images , and clinical images of the skin The upper left corner is the origin of the coordinate system;

[0086] Record all local small images The coordinates of the upper left corner in the coordinate system, and the coordinates are calculated to make Small local images cover the skin clinical image All pixel information in the image, the coordinate calculation is as follows:

[0087] , ∈Z and ;

[0088] in, represents the number of local small images, H represents the side length of the skin clinical image, h represents the side length of the local small image, Indicates rounding up;

[0089] Construct a sampling data set based on the local small image :

[0090] ;

[0091] in, The coordinate information of each local small image is recorded in.

[0092] In the above embodiment, several clinical images of skin of different parts of the human body are collected, including all parts of the human body such as face, head, neck, chest, back, etc. The collected images cover all age groups and have a balanced gender ratio, thereby ensuring the diversity of the data and forming a data set. ,in and Respectively represent clinical images and their corresponding part labels, is the sequence number of the image, is the number of images.

[0093] set up The pixel size is ,from Random non-overlapping sampling is performed in Random Intercept The pixel size is A small local image of The upper left corner of is used as the origin of the coordinate system to record the coordinates of the upper left corner of each local small image, as shown in the following figure; , ∈Z and , Indicates rounding up to achieve the intercepted Get the covered image by using local small images All pixel information.

[0094] Perform random sampling as described in step 1.2 on all images in the dataset P to form a dataset ,in The coordinate information of each local small image is recorded in.

[0095] Build a human body part recognition model based on convolutional neural network and use the dataset Perform end-to-end training until the model converges and inputs small local images , output the human body part label to which the small image belongs .

[0096] Build an unsupervised super-resolution magnification model based on convolutional neural network and use the dataset The local small images in the training super-resolution enlargement model are trained in an unpaired manner (without using super-resolution images) until the model converges and the input local small images are realized. , output the super-resolution enlarged image of the small image .

[0097] Optionally, selecting a local small image similar to the skin tumor resection margin image to be designed in the sampled data set by similarity comparison includes:

[0098] The human body part labels in the skin tumor resection edge image to be designed are used to compare the phase velocity with the local small image, and the structural similarity, Pearson correlation coefficient and Manhattan distance between the skin tumor resection edge image to be designed and the local small image are calculated;

[0099] The weighted sum of the calculation results is performed to obtain the similarity value between the skin tumor resection edge image to be designed and the local small image. The calculation formula is as follows:

[0100] ;

[0101] in, Indicates structural similarity, represents the Pearson correlation coefficient, represents the Manhattan distance, , , are fitting parameters.

[0102] Optionally, using the coordinate information of the local small image to solve the reference coordinate range of the skin tumor resection margin image to be designed includes:

[0103] The coordinate information of the local small image is extracted, and the mean value and standard deviation of the coordinate information are solved. Based on the mean value and standard deviation of the coordinate information, the reference coordinate range of the skin tumor resection edge image to be designed is obtained: ;

[0104] in, Represents the average value of coordinate information, Indicates the standard deviation of the coordinate information.

[0105] Optionally, the performing skin texture detection on the super-resolution magnified image, thereby achieving self-correction of rough skin grain lines to obtain accurate skin grain lines, includes:

[0106] A random seed point is set on the rough skin grain line in the super-resolution magnified image, and the adjacent pixel points within 360° of the seed point are used as the search range. The similarity between the random seed point and the adjacent pixel points in the search range is calculated as shown in the following formula:

[0107]

[0108] in, , as well as and , as well as are the three color channel values ​​of the searched pixel and seed point in the CIELab color space respectively;

[0109] Taking any two adjacent pixels in the search range as comparison units, calculate the similarity values ​​between the adjacent pixels in the comparison unit and the random seed points, and sum them up as the similarity value of the comparison unit;

[0110] Calculate the distances between two adjacent pixels in the comparison unit and other pixels on the rough dermatoglyphic line within the search range that are closest to the comparison unit pixel, excluding the seed pixel, and sum them up as the distance measurement value of the comparison unit;

[0111] The self-correction metric value of the comparison unit is obtained by adding the similarity value of the comparison unit to the inverse of the distance metric value, the self-correction metric values ​​of all comparison units in the search range are calculated, and the sizes are compared, and the seed point is updated to the comparison unit with the largest self-correction metric value;

[0112] Repeat the above steps until the random seed points are on both sides of the rough dermatoglyphics, and use the movement trajectory of the random seed points in the self-correction process as the precise dermatoglyphics line.

[0113] In the above embodiment, for any local small image of the skin tumor resection margin to be designed, First, it inputs the trained human body part recognition model to obtain a small local image Human body part label .

[0114] Use small local images With the dataset Human body part labels and All the same local small images are compared for similarity, and the structural similarity between them is calculated. , Pearson correlation coefficient , Manhattan distance Three groups of indicators are weighted and summed to obtain the local small image and dataset The similarity value between local small images with the same human body parts :

[0115] .

[0116] The dataset Neutralize small local images Similarity value between Highest The coordinate information of the local small image is taken out, and the obtained Coordinate information to solve the average value and standard deviation , get a small local image Reference coordinate range .

[0117] Based on the local small image Reference coordinate range , compared with human body parts Human dermatoglyphics atlas, obtain local small images The rough direction of the skin grain.

[0118] The local small image Input the trained super-resolution magnification model to obtain the super-resolution magnified image ; For super-resolution magnified images Detect skin texture and realize the acquisition of local small images The rough dermatoglyphic lines are self-corrected to obtain more precise dermatoglyphic lines, specifically:

[0119] In super-resolution magnified image A random seed point is set for each rough leather grain line in pixels, The pixel width of the skin grain line is calculated by taking the set random seed point as the search starting point and calculating the similarity between the seed point and the adjacent pixel points in the 360° direction. , as shown below:

[0120] ;

[0121] in, , as well as and , as well as are the three color channel values ​​of the searched pixel and seed point in the CIELab color space respectively.

[0122] Take any two adjacent pixels in the search range as comparison units and calculate the Pixels and seed points The similarity between pixels is summed as the similarity value of the comparison unit. ; Calculate the distance between the two pixels in the comparison unit and the other pixels closest to the comparison unit pixel on the rough skin line within the search range, except the seed pixel, and sum them as the distance measurement value of the comparison unit .

[0123] The similarity values ​​of the compared units The distance metric between the comparison unit and The reciprocal of is added to obtain the self-correction metric value of the comparison unit ( ); Calculate the self-correction metric values ​​of all comparison units within the search range, and update the seed point to the comparison unit with the largest self-correction metric value.

[0124] Repeat the above steps until the seed point reaches both sides of the rough dermatoglyphic line; the movement trajectory of the seed point during the self-correction process is used as the corrected dermatoglyphic line, that is, the precise dermatoglyphic line.

[0125] Optionally, the performing resection margin design on the skin tumor in the image of the skin tumor resection margin to be designed based on the tumor boundary information and the precise skin grain line includes:

[0126] The tumor center was taken as the original resection center, and the resection margin was designed according to the precise skin grain line direction. The resection margin design was then corrected secondary based on the patient's gender.

[0127] In the above embodiment, for any local small image of the skin tumor resection margin to be designed, , for a small local image The tumor boundary is detected, including but not limited to boundary detection methods based on deep learning, Canny edge detectors in traditional image processing methods, Harris corner detectors and other methods.

[0128] Based on local small image The tumor boundary and precise skin lines are used to generate the preliminary cutting margin design of the skin tumor, that is, the cutting margin is designed with the tumor center as the cutting margin center and the precise skin lines as the parallel line of the cutting margin. Figure 6 shown.

[0129] The elasticity of human skin gradually decreases with age, and the elasticity of male skin is higher than that of female skin. Therefore, the initial cutting edge design is further corrected according to the patient's gender and age information, specifically:

[0130] If the patient is male and is , then the secondary correction margin width is:

[0131] ;

[0132] In the formula, Indicates the width of the secondary correction margin in male patients. Indicates the age of male patients, Indicates the width of the cutting edge;

[0133] If the patient is female and is , then the secondary correction margin width is:

[0134] ;

[0135] Indicates the width of the secondary correction margin in female patients. Indicates the age of female patients, Indicates the width of the cutting edge.

[0136] In the above embodiment, for any local small image of the skin tumor resection margin to be designed, , for a small local image The tumor boundary is detected, including but not limited to boundary detection methods based on deep learning, Canny edge detectors in traditional image processing methods, Harris corner detectors and other methods.

[0137] Based on local small image The tumor boundary and precise skin lines are used to generate the preliminary cutting margin design of the skin tumor, that is, the cutting margin is designed with the tumor center as the cutting margin center and the precise skin lines as the parallel line of the cutting margin. Figure 6 shown.

[0138] The elasticity of human skin gradually decreases with age, and the elasticity of male skin is higher than that of female skin. Therefore, the initial cutting edge design is further corrected according to the patient's gender and age information, specifically:

[0139] If the patient is male and is , then the secondary correction margin width is:

[0140] ;

[0141] In the formula, Indicates the width of the secondary correction margin in male patients. Indicates the age of male patients, Indicates the width of the cutting edge;

[0142] If the patient is female and is , then the secondary correction margin width is:

[0143] ;

[0144] Indicates the width of the secondary correction margin in female patients. Indicates the age of female patients, Indicates the width of the cutting edge.

[0145] The embodiment of the present application also provides a tumor margin design system based on a local skin image, including a processor and a memory;

[0146] Memory, used to store computer programs;

[0147] The processor is used to implement any method step in the method for designing tumor resection margin based on local skin image when executing the program stored in the memory.

[0148] The above-mentioned tumor resection margin design system based on local skin images can implement various embodiments of the above-mentioned tumor resection margin design method based on local skin images, and can achieve the same beneficial effects, which will not be described in detail here.

[0149] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A tumor resection margin design method based on local skin images, characterized in that: include: Acquire clinical skin images of different parts of the human body and accurate part labels corresponding to the clinical skin images to construct an initial data set, and perform random non-overlapping sampling on the initial data set to obtain a sampled data set; Constructing a super-resolution magnification model and a human body part recognition model based on the sampled data set; Acquire a small local image of the skin tumor resection margin to be designed, and input the small local image of the skin tumor resection margin to be designed into the human body part recognition model to obtain a human body part label; Based on the human body part label, a similarity comparison is performed between the local small image of the skin tumor resection margin to be designed and the local small images in the sampling data set, and the reference coordinate range of the local small image of the skin tumor resection margin to be designed is solved by using the coordinate information of the local small image with high similarity in the sampling data set; Obtaining a rough skin grain line of a local small image of a skin tumor resection margin to be designed based on the human body part label and the reference coordinate range; Inputting a local small image containing rough skin grain lines into the super-resolution magnification model to obtain a super-resolution magnified image containing rough skin grain lines, and performing skin texture detection on the super-resolution magnified image containing rough skin grain lines to obtain precise skin grain lines; Tumor boundary detection is performed on the local small image of the skin tumor resection margin to be designed to obtain tumor boundary information, and the skin tumor resection margin is designed in the local small image of the skin tumor resection margin to be designed based on the tumor boundary information and the precise skin grain line.

2. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: The method of constructing a super-resolution magnification model and a human body part recognition model based on a sampled data set includes: A super-resolution magnification model and a human body part recognition model based on a convolutional neural network are constructed, and the super-resolution magnification model and the human body part recognition model are trained respectively using the sampled data set to obtain a trained super-resolution magnification model and a trained human body part recognition model.

3. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: The initial data set is constructed by obtaining clinical skin images of different parts of the human body and accurate part labels corresponding to the clinical skin images. ,include: Obtain skin clinical images of different parts of the human body and the accurate part labels corresponding to the skin clinical images to build the initial data set , where accurate part labels include: face, head, neck, chest, and back; The initial dataset is: ； in, represents the initial data set, and They represent the skin clinical image and the part label corresponding to the skin clinical image, is the sequence number of the image, is the number of images.

4. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: The initial dataset Clinical images of the skin in Perform random non-overlapping sampling to obtain the sampled data set ,include: Clinical images in skin Random Intercept Small local images , and clinical images of the skin The upper left corner is the origin of the coordinate system; Record all local small images The coordinates of the upper left corner in the coordinate system, and the coordinates are calculated to make Small local images Overlay skin clinical images All pixel information in the image, the coordinate calculation is as follows: , ∈Z and ; in, represents the number of local small images, H represents the side length of the skin clinical image, h represents the side length of the local small image, Indicates rounding up. Indicates the number of samples; Construct a sampling data set based on the local small image : ； in, represents the sampled data set, The coordinate information of each local small image is recorded in represents a small local image, Indicates the part label corresponding to the skin clinical image, is the number of images.

5. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: The method of using the coordinate information of the local small images with high similarity in the sampled data set to solve the reference coordinate range of the local small image of the skin tumor resection margin to be designed includes: Calculate the local small image of the skin tumor resection margin to be designed With the sampled data set Zhongyu Small local images with the same human body part labels structural similarity, Pearson correlation coefficient and Manhattan distance; The weighted sum of the calculation results is used to obtain a small local image of the skin tumor margin to be designed. With local small image The similarity value between , and its calculation formula is as follows: ； in, represents the similarity value, Indicates structural similarity, represents the Pearson correlation coefficient, represents the Manhattan distance, , , are fitting parameters.

6. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: Using a small local image of the skin tumor margin to be designed With local small image The similarity information between them is used to solve the local small image of the skin tumor resection margin to be designed The reference coordinate range includes: Extracting the dataset Small local image of the cutaneous tumor margin to be designed Similarity value between The highest front Small local images The coordinate information of the skin tumor is obtained by solving the mean value and standard deviation of the coordinate information, and the local small image of the skin tumor resection margin to be designed is obtained based on the mean value and standard deviation of the coordinate information. Reference coordinate range: ; in, Represents the average value of coordinate information, Indicates the standard deviation of the coordinate information.

7. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: The method of obtaining a rough skin grain line of a local small image of a skin tumor resection margin to be designed based on the human body part label and the reference coordinate range includes: Using a small local image of the skin tumor margin to be designed Reference coordinate range , compare the human dermatoglyphics atlas of the corresponding human body parts, and obtain the local small image of the skin tumor resection margin to be designed The direction of the rough skin grain; in, Represents the average value of coordinate information, Indicates the standard deviation of the coordinate information.

8. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: The step of performing skin texture detection on the super-resolution magnified image containing rough skin grain lines to obtain accurate skin grain lines comprises: A random seed point is set on the rough skin grain line in the super-resolution magnified image, and the adjacent pixel points within 360° of the seed point are used as the search range. The similarity between the random seed point and the adjacent pixel points in the search range is calculated as shown in the following formula: in, represents the similarity value, , as well as and , as well as are the three color channel values ​​of the searched pixel and seed point in the CIELab color space respectively; Taking any two adjacent pixels in the search range as comparison units, calculate the similarity values ​​between the adjacent pixels in the comparison unit and the random seed points, and sum them up as the similarity value of the comparison unit; Calculate the distances between two adjacent pixels in the comparison unit and other pixels on the rough dermatoglyphic line within the search range that are closest to the comparison unit pixel, excluding the seed pixel, and sum them up as the distance measurement value of the comparison unit; The self-correction metric value of the comparison unit is obtained by adding the similarity value of the comparison unit to the inverse of the distance metric value, the self-correction metric values ​​of all comparison units in the search range are calculated, and the sizes are compared, and the seed point is updated to the comparison unit with the largest self-correction metric value; Repeat the above steps until the random seed points are on both sides of the rough dermatoglyphics, and use the movement trajectory of the random seed points in the self-correction process as the precise dermatoglyphics line.

9. The tumor resection margin design method based on skin local image according to claim 1, characterized in that: The step of designing the skin tumor resection margin in the local small image for designing the skin tumor resection margin based on the tumor boundary information and the precise skin grain line comprises: The tumor center was used as the cutting edge center, and the cutting edge was designed according to the precise skin grain line direction. The cutting edge design was corrected again based on the patient's gender. If the patient is male and is , then the secondary correction margin width is: ； In the formula, Indicates the width of the secondary correction margin in male patients. Indicates the age of male patients, Indicates the width of the cutting edge; If the patient is female and is , then the secondary correction margin width is: ； Indicates the width of the secondary correction margin in female patients. Indicates the age of female patients, Indicates the width of the cutting edge.

10. A tumor margin design system based on local skin images, characterized in that: Including processor and memory; Memory, used to store computer programs; A processor, for implementing the method steps described in any one of claims 1 to 9 when executing a program stored in a memory.

Images