CN117541580A - A method for establishing a thyroid cancer image comparison model based on deep neural network - Google Patents

Abstract

The invention discloses a thyroid cancer image comparison model establishment method based on a deep neural network, which relates to the technical field of image recognition and comprises the following steps: acquiring a plurality of training images, and marking thyroid cancer areas in the training images, wherein the training images comprise thyroid cancer areas; extracting gray features of the thyroid region, and training gray features of the thyroid region of the training images to obtain gray comparison parameters; dividing the thyroid region into an integral region and a punctiform region, and performing shape feature training on the integral region and the punctiform region of a plurality of training images to obtain shape comparison parameters; the method is used for solving the problem that the existing thyroid cancer image identification method lacks a specific feature extraction means, so that effective feature comparison identification cannot be performed.

Description

A method for establishing a thyroid cancer image comparison model based on deep neural network

Technical field

The present invention relates to the field of image recognition technology, specifically a method for establishing a thyroid cancer image comparison model based on a deep neural network.

Background technique

Image recognition refers to the technology of using computers to process, analyze and understand images to identify various targets and objects of different modes. It is a practical application of deep learning algorithms. Usually, the image recognition process is divided into four There are three steps: image acquisition, image preprocessing, feature extraction and image recognition. Deep neural network is a technology in the field of machine learning. In the process of image recognition, the deep neural network is used to extract features from training samples, which can improve the image quality. Identify the accuracy of the comparison.

In the existing technology, in the process of identifying thyroid cancer images, image recognition technology is used for feature extraction and comparison. For example, the Chinese patent application with publication number CN112233106A discloses a residual capsule network-based An ultrasonic image analysis method for thyroid cancer. This method analyzes the ultrasonic images of thyroid cancer through a residual capsule network to obtain the ultrasonic image classification and recognition results corresponding to the ultrasonic image of papillary thyroid cancer to be identified. However, this method only discloses the residual Capsule network is a technology for analyzing and classifying images. In step S1 of the method, each image in the original papillary thyroid cancer ultrasound image data set includes irregular shape attributes, unclear boundary attributes, uneven echo attributes, and calcification attributes. and one or more of the normal attributes. However, this method only lists the above-mentioned features in thyroid cancer images and lacks a specific identification scheme for the above-mentioned features. This method cannot be used to effectively compare and identify thyroid cancer images. Therefore, a method that can effectively extract and compare features in thyroid cancer images is needed to solve the above problems.

Contents of the invention

The present invention aims to solve one of the technical problems in the prior art at least to a certain extent. By extracting features from several thyroid cancer images and establishing an image comparison model based on the extracted features, it can help improve the feature ratio of image screening. Regarding accuracy, it solves the problem that existing thyroid cancer image recognition methods lack specific feature extraction means, resulting in the inability to carry out effective feature comparison and recognition.

In order to achieve the above purpose, the first aspect of this application provides a method for establishing a thyroid cancer image comparison model based on a deep neural network, which includes: acquiring a number of training images, and marking the thyroid cancer areas in the training images. Includes areas of thyroid cancer;

Extract grayscale features from the thyroid area, perform grayscale feature training on the thyroid area of several training images, and obtain grayscale comparison parameters;

Divide the thyroid area into overall areas and point-like areas, conduct shape feature training on the overall areas and point-like areas of several training images, and obtain shape comparison parameters;

An image comparison model is established based on grayscale comparison parameters and shape comparison parameters.

Further, obtaining several training images and marking the thyroid cancer area in the training images includes: dividing the training images into pixels and establishing a two-dimensional coordinate system based on the pixels;

The pixels in the thyroid cancer area are marked with coordinates in a two-dimensional coordinate system, and the pixels in the thyroid cancer area are set as comparison pixels.

Further, grayscale features are extracted from the thyroid area, and grayscale feature training is performed on the thyroid area of several training images. The grayscale comparison parameters obtained include: setting the pixel points other than the comparison pixel points in the two-dimensional coordinate system. Defined as peripheral pixels;

Set the pixel points adjacent to the comparison pixel point and the peripheral pixel point as the comparison outline pixel point, and set the pixel points adjacent to the peripheral pixel point and the comparison outline pixel point as the peripheral contact pixel point;

Calculate the average gray level of the comparison contour pixels in the training image, and set it as the comparison contour gray level; calculate the average gray level of the peripheral connecting pixels in the training image, and set it as the peripheral connecting gray level. degree; calculate the absolute value of the difference between the contrasting contour grayscale and the peripheral grayscale, and set it as the grayscale comparison value;

Perform grayscale feature training on several training images one by one through the above steps to obtain several comparison contour grayscales, several peripheral connecting grayscales, and several grayscale comparison values;

Several comparison contour grayscales, several peripheral connecting grayscales, and several grayscale comparison values are processed through a comparison parameter extraction method to obtain grayscale comparison parameters. The grayscale comparison parameters include the comparison contour grayscale range. , peripheral connected grayscale range and grayscale comparison value range.

Further, the comparison parameter extraction method includes: obtaining the maximum value and the minimum value of a set of input grayscale values, and setting them as the maximum grayscale value and the minimum grayscale value respectively; a set of grayscale values includes several comparisons. A group of contour grayscales, several peripheral grayscales, or several grayscale comparison values;

Subtract the minimum gray value from the maximum gray value to obtain the gray difference, divide the gray difference by the first preset number to obtain the first undetermined division value, extract the integer bit of the first undetermined division value and add one to obtain the second Pending partition value;

Use the minimum gray value as the starting point for division, use the second undetermined division value as the division unit, divide a first preset number of gray scale groups, correspond the input set of gray scale values to each gray scale group, and select the gray scale The gray level group with the largest number of distribution values is set as the gray level comparison parameter group;

Set the grayscale range of the grayscale comparison parameter group as the grayscale comparison parameter.

Further, dividing the thyroid area into an overall area and a point area includes: obtaining the comparison pixel points in the two-dimensional coordinate system, and setting the area of interconnected comparison pixel points as the overall undetermined area;

Obtain the number of the overall undetermined area in the two-dimensional coordinate system. When the number of the overall undetermined area is less than or equal to the first distribution quantity threshold, the overall undetermined area is set as the overall area. When the number of the overall undetermined area is greater than the first distribution quantity threshold, , setting the overall undetermined area as a point area.

Further, shape feature training is performed on the entire area of several training images, and the obtained shape comparison parameters include: setting a peripheral circle to frame the entire area, and the outer circle is the smallest circle that can completely frame the entire area;

Each time the radius of the outer circle is reduced by the first unit length, an updated circle is obtained, and the center of the updated circle is consistent with the center of the outer circle;

The overall area inside the update circle obtained each time is set as the internal area to be divided, and the overall area between the update circle obtained each time and the adjacent update circle or peripheral circle outside is set as the cutting area;

Set the area of interconnected comparison pixels in the cutting area as cutting independent areas, count the number of cutting independent areas in each group of cutting areas, and set it as the edge divergence number; when the edge divergence number is less than or equal to the first independent number When the threshold is reached, stop reducing the radius of the update circle or peripheral circle;

Obtain the maximum value of the number of edge divergences of several groups and set it as the number of overall divergence distribution;

The overall divergence distribution quantity corresponding to several training images is processed through the shape comparison extraction method to obtain the overall comparison parameter, and the overall comparison parameter includes the overall divergence distribution quantity range.

Further, shape feature training is performed on the point-like areas of several training images, and the obtained shape comparison parameters include: obtaining the number of point-like areas in the training image and setting it as the number of point-like distributions;

The number of point distributions corresponding to several training images is processed through a shape comparison extraction method to obtain point comparison parameters, where the point comparison parameters include a range of point distribution numbers.

Further, the shape comparison extraction method includes: obtaining the maximum value and the minimum value of a set of input distribution amounts, and setting them as the maximum value and the minimum distribution amount respectively; a set of distribution amounts includes a plurality of training images corresponding to The number of overall divergent distributions or a group of the number of point distributions corresponding to several training images;

Subtract the minimum value of the distribution amount from the maximum value of the distribution amount to obtain the distribution amount difference, divide the distribution amount difference by the second preset number to obtain the first distribution amount division value, extract the integer digit of the first distribution amount division value and add one to obtain The second distribution quantity division value;

Use the minimum value of the distribution amount as the starting point for division, use the second distribution amount division value as the division unit, divide the second preset number of distribution amount groups, correspond the input set of distribution amounts to each distribution amount group, and select the distribution The distribution quantity group with the largest quantity distribution is set as the distribution quantity comparison parameter group;

The range of the distribution comparison parameter group is set as a shape comparison parameter, and the shape comparison parameter includes an overall comparison parameter and a point comparison parameter.

Beneficial effects of the present invention: By acquiring several training images and marking the thyroid cancer area in the training images, the present invention can facilitate accurate regional correspondence during feature training and improve the accuracy of data acquisition; by performing grayscale features on the thyroid area Extract, conduct grayscale feature training on the thyroid area of several training images, and obtain grayscale comparison parameters. Through the grayscale comparison parameters, a good feature comparison framework can be built in the image comparison model, and the features can be improved for preliminary comparison and identification. efficiency and effectiveness;

The present invention divides the thyroid area into overall areas and point-like areas, performs shape feature training on the overall areas and point-like areas of several training images, and obtains shape comparison parameters, and then uses the shape comparison parameters to obtain the results obtained through grayscale screening. The feature comparison area is further characterized by feature comparison and screening, thereby improving the effectiveness and accuracy of thyroid cancer image comparison and recognition; finally, an image comparison model is established based on grayscale comparison parameters and shape comparison parameters, and the image comparison model is based on The establishment of grayscale comparison parameters and shape comparison parameters can improve the accuracy of image comparison in practical applications.

Additional features and advantages of the application will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Description of drawings

Figure 1 is a step flow chart of the method of the present invention;

Figure 2 is a schematic diagram of obtaining an independent cutting area according to the present invention;

Figure 3 is a schematic diagram of the entire area included in the training image of the present invention;

Figure 4 is a schematic diagram of a dotted area included in the training image of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Embodiment 1 As shown in Figure 1, a method for establishing a thyroid cancer image comparison model based on a deep neural network is used to extract features from several thyroid cancer images and establish an image comparison model based on the extracted features, which can help Improve the feature comparison accuracy of image screening to solve the problem that existing thyroid cancer image recognition methods lack specific feature extraction methods, resulting in the inability to perform effective feature comparison and recognition.

Specifically, the method for establishing a thyroid cancer image comparison model based on a deep neural network includes the following steps: Step S1, obtain a number of training images, mark the thyroid cancer area in the training image, and the training image includes the thyroid cancer area; Step S1 also It includes the following sub-steps: Step S101, divide the training image into pixels, and establish a two-dimensional coordinate system based on the pixels. During specific implementation, the pixels are divided according to a width of 1280 pixels and a height of 720 pixels;

Step S102: Mark the pixels in the thyroid cancer area with coordinates in a two-dimensional coordinate system, and set the pixels in the thyroid cancer area as comparison pixels.

Step S2: Extract grayscale features from the thyroid region, perform grayscale feature training on the thyroid region of several training images, and obtain grayscale comparison parameters; Step S2 also includes: Step S2011, compare the pixels in the two-dimensional coordinate system Pixels outside the point are set as peripheral pixels;

Step S2012, set the pixel points adjacent to the comparison pixel point and the peripheral pixel point as the comparison outline pixel point, and set the pixel points adjacent to the peripheral pixel point and the comparison outline pixel point as the peripheral contact pixel point;

Step S2013: Calculate the average grayscale of the comparison contour pixels in the training image and set it as the comparison contour grayscale; calculate the average grayscale of the peripheral adjacent pixels in the training image and set it as the peripheral grayscale. Connected grayscale; calculate the absolute value of the difference between the contrasting contour grayscale and the peripheral connected grayscale, and set it as the grayscale comparison value;

Step S2014, perform grayscale feature training on several training images one by one through steps S2011 to S2013 to obtain several comparison contour grayscales, several peripheral connection grayscales, and several grayscale comparison values;

Step S2015: A plurality of comparison contour grayscales, a plurality of peripheral connecting grayscales and a plurality of grayscale comparison values are respectively processed through a comparison parameter extraction method to obtain grayscale comparison parameters. The grayscale comparison parameters include comparison contour grayscales. range, peripheral connected grayscale range and grayscale comparison value range. The obtained comparison contour grayscale range, peripheral connected grayscale range and grayscale comparison value range can facilitate grayscale comparison during actual image comparison. The degree comparison parameter delimits the initial feature area, which helps to improve the efficiency of preliminary feature extraction, thereby ensuring that the data processing volume of the next step of shape comparison is reduced while improving the accuracy of data comparison.

The comparison parameter extraction method includes the following steps: Step S2021, obtain the maximum value and minimum value of a set of input grayscale values, and set them as the maximum grayscale value and the minimum grayscale value respectively; a set of grayscale values includes several comparisons A group of contour grayscales, several peripheral grayscales, or several grayscale comparison values;

Step S2022: Subtract the minimum gray value from the maximum gray value to obtain the gray difference, divide the gray difference by the first preset number to obtain the first undetermined division value, and extract the integer digit of the first undetermined division value plus one. Obtain the second undetermined division value; the first preset number is set to 10, for example, when the grayscale difference is 55, the first undetermined division value obtained is 5.5, and the second undetermined division value is 6;

Step S2023, use the minimum gray value as the starting point for division, use the second undetermined division value as the division unit, divide the first preset number of gray scale groups, and correspond the input set of gray scale values to each gray scale group. , select the grayscale group with the largest number of grayscale values and set it as the grayscale comparison parameter group;

Step S2024: Set the grayscale range of the grayscale comparison parameter group as the grayscale comparison parameter.

Please refer to Figures 2 to 4, step S3, divide the thyroid area into an overall area and a point area. In Figure 3, the area pointed by the gray arrow is the overall area, and the area pointed by the white arrow in Figure 4 is a point. Shape area, perform shape feature training on the overall area and point area of several training images, and obtain shape comparison parameters; Step S3 also includes: Step S3011, obtain the comparison pixel points in the two-dimensional coordinate system, and compare the connected comparison pixels. Set the area of pixels as the overall undetermined area;

Step S3012, obtain the number of the overall undetermined areas in the two-dimensional coordinate system. When the number of the overall undetermined areas is less than or equal to the first distribution quantity threshold, set the overall undetermined area as the overall area. When the number of the overall undetermined areas is greater than the first distribution When setting the quantity threshold, the entire undetermined area is set to a point-like area, and the first distribution quantity is set to 3. Normally, the entire area is connected together, so the value of the first distribution quantity does not need to be set too large.

Step S3 also includes: step S3021, setting a peripheral circle to frame the entire area, and the outer circle is the smallest circle that can completely frame the entire area;

Step S3022, each time the radius of the peripheral circle is reduced by the first unit length to obtain an update circle, the center of the update circle is consistent with the center of the peripheral circle; the first unit length is set according to the side length of the pixel, as shown in Figure 2 , the radius difference between the update circle and the peripheral circle is one pixel side length, and the first unit length is specifically set to the side length of one pixel point;

Step S3023, set the entire area inside the update circle obtained each time as the internal area to be divided, and set the entire area between the update circle obtained each time and the external adjacent update circle or peripheral circle as the cutting area;

Step S3024: Set the area of interconnected comparison pixels in the cutting area as cutting independent areas, count the number of cutting independent areas in each group of cutting areas, and set it as the number of edge divergences; when the number of edge divergences is less than or equal to the th. When an independent quantity threshold is reached, stop reducing the radius of the update circle or peripheral circle; when the first independent threshold is set to 3, when the number of edge divergences is less than or equal to 3, stop reducing the update circle or peripheral circle;

Step S3025, obtain the maximum value of several groups of edge divergence numbers, and set it as the overall divergence distribution number; when obtaining the edge divergence number, the greater the edge divergence number, the more irregular the edges of the overall area, acupuncture-like or other-shaped edges. The more structure there is;

Step S3026: The overall divergence distribution quantity corresponding to several training images is processed through the shape comparison extraction method to obtain the overall comparison parameter. The overall comparison parameter includes the overall divergence distribution quantity range.

Step S3 also includes: step S3031, obtaining the number of point-like areas in the training image and setting it as the number of point-like distributions; the greater the number of point-like distributions, the more calcification points are in the training image;

Step S3032: Process the number of point distributions corresponding to several training images through the shape comparison extraction method to obtain point comparison parameters. The point comparison parameters include the range of the number of point distributions.

The shape comparison extraction method includes: step S3041, obtaining the maximum value and the minimum value of a set of input distribution amounts, and setting them as the maximum value and the minimum distribution amount respectively; a set of distribution amounts includes the overall divergence corresponding to several training images The number of distributions or a group of the number of point distributions corresponding to several training images;

Step S3042: Subtract the minimum value of the distribution amount from the maximum value of the distribution amount to obtain the distribution amount difference, divide the distribution amount difference by the second preset number to obtain the first distribution amount division value, and extract the integer digits of the first distribution amount division value. Add one to get the second distribution amount division value; the second preset number is specifically set to 5. For example, when the distribution amount difference is 99, the obtained first distribution amount division value is 19.8, and the second distribution amount division value is 20;

Step S3043, use the minimum value of the distribution amount as the starting point for division, use the second distribution amount division value as the division unit, divide the second preset number of distribution amount groups, and correspond the input set of distribution amounts to each distribution amount group. , select the distribution group with the largest number of distributions and set it as the distribution comparison parameter group;

Step S3044 sets the range of the distribution comparison parameter group as the shape comparison parameter, and the shape comparison parameter includes the overall comparison parameter and the point comparison parameter.

Step S4, establish an image comparison model based on grayscale comparison parameters and shape comparison parameters; during specific implementation, the image that needs to be compared and identified is input into the image comparison model. The processing process of the image comparison model includes: first, input The images are compared through the comparison contour grayscale range, peripheral adjacent grayscale range and grayscale comparison value range in the grayscale comparison parameters, and the area that needs to be initially identified is divided and set as the area to be identified. If the area to be identified is not extracted through the grayscale comparison parameters, an unidentified feature signal is output. The unidentified feature signal indicates that no features similar to the parameters in the image comparison model are extracted from the input image; then the area to be identified is The shape is compared, and the shape division type of the area to be identified is obtained by comparing it with the overall comparison parameter and the point comparison parameter in the shape comparison parameters. If the area to be identified is compared with the overall shape, If there is no similarity between the parameters and the point comparison parameters, a pending signal with no identification features will be output. The pending signal with no identification features indicates that there may be a risk area in the image, and further manual verification is required.

In the second aspect of Embodiment 2, the present application provides an electronic device, including a processor and a memory. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, any one of the above methods is executed. step. Through the above technical solution, the processor and the memory are interconnected and communicate with each other through a communication bus and/or other forms of connection mechanisms. The memory stores a computer program executable by the processor. When the electronic device is running, the processor executes the computer program. The method in any optional implementation of the above embodiment is executed during execution to achieve the following functions: first obtain several training images, mark the thyroid cancer area in the training images; and then extract grayscale features from the thyroid area , perform grayscale feature training on the thyroid area of several training images to obtain grayscale comparison parameters; then divide the thyroid area into overall areas and point-like areas, and perform shape feature training on the overall areas and point-like areas of several training images. The shape comparison parameters are obtained; finally, an image comparison model is established based on the grayscale comparison parameters and shape comparison parameters.

In the third aspect of the third embodiment, the present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in any one of the above methods are executed. Through the above technical solution, when the computer program is executed by the processor, the method in any optional implementation of the above embodiment is executed to achieve the following functions: first, obtain several training images, and mark the thyroid cancer area in the training images. ; Then perform grayscale feature extraction on the thyroid area, perform grayscale feature training on the thyroid area of several training images, and obtain grayscale comparison parameters; then divide the thyroid area into overall areas and point-like areas, and perform overall training on several training images. Conduct shape feature training on regions and point-like regions to obtain shape comparison parameters; finally, an image comparison model is established based on grayscale comparison parameters and shape comparison parameters.

Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems or computer program products. Thus, the invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the invention may take the form of a computer program product embodied on one or more computer-usable storage media embodying computer-usable program code therein. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or their combination, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (electrically erasable programmable read-only memory) Electrically Erasable ProgrammableRead-Only Memory (EEPROM for short), Erasable ProgrammableRead Only Memory (EPROM for short), Programmable Red-Only Memory (PROM for short), Read-only memory (Read -OnlyMemory (ROM for short), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions The device implements the functions specified in a process or processes in the flowchart and/or in a block or blocks in the block diagram.

In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or can be integrated into another system, or some features can be ignored, or not implemented. Another point is that the coupling or direct coupling or communication connection between each other shown or discussed may be through some communication interfaces, and the indirect coupling or communication connection of the devices or units may be in electrical, mechanical or other forms.

Claims (8)

1. A method for establishing a thyroid cancer image comparison model based on a deep neural network, which is characterized in that it includes: acquiring a number of training images, and marking the thyroid cancer area in the training image, and the training image includes the thyroid cancer area; Extract grayscale features from the thyroid area, perform grayscale feature training on the thyroid area of several training images, and obtain grayscale comparison parameters; Divide the thyroid area into overall areas and point-like areas, conduct shape feature training on the overall areas and point-like areas of several training images, and obtain shape comparison parameters; An image comparison model is established based on grayscale comparison parameters and shape comparison parameters. 2. A method for establishing a thyroid cancer image comparison model based on a deep neural network according to claim 1, characterized in that, obtaining a number of training images, and marking the thyroid cancer area in the training images includes: Pixel point division, establishing a two-dimensional coordinate system based on pixel points; The pixels in the thyroid cancer area are marked with coordinates in a two-dimensional coordinate system, and the pixels in the thyroid cancer area are set as comparison pixels. 3. A method for establishing a thyroid cancer image comparison model based on a deep neural network according to claim 2, characterized in that grayscale feature extraction is performed on the thyroid region, and grayscale feature training is performed on the thyroid region of several training images. , obtaining the grayscale comparison parameters includes: setting pixels outside the comparison pixels as peripheral pixels in the two-dimensional coordinate system; Set the pixel points adjacent to the comparison pixel point and the peripheral pixel point as the comparison outline pixel point, and set the pixel points adjacent to the peripheral pixel point and the comparison outline pixel point as the peripheral contact pixel point; Calculate the average gray level of the comparison contour pixels in the training image, and set it as the comparison contour gray level; calculate the average gray level of the peripheral connecting pixels in the training image, and set it as the peripheral connecting gray level. degree; calculate the absolute value of the difference between the contrasting contour grayscale and the peripheral grayscale, and set it as the grayscale comparison value; Perform grayscale feature training on several training images one by one through the above steps to obtain several comparison contour grayscales, several peripheral connecting grayscales, and several grayscale comparison values; Several comparison contour grayscales, several peripheral connecting grayscales, and several grayscale comparison values are processed through a comparison parameter extraction method to obtain grayscale comparison parameters. The grayscale comparison parameters include the comparison contour grayscale range. , peripheral connected grayscale range and grayscale comparison value range. 4. A method for establishing a thyroid cancer image comparison model based on a deep neural network according to claim 3, characterized in that the comparison parameter extraction method includes: obtaining the maximum value and sum of a set of input grayscale values. The minimum value is set to the maximum gray value and the minimum gray value respectively; a group of gray values includes a group of several comparison contour gray levels, several peripheral connecting gray levels, or several gray level comparison values; Subtract the minimum gray value from the maximum gray value to obtain the gray difference, divide the gray difference by the first preset number to obtain the first undetermined division value, extract the integer bit of the first undetermined division value and add one to obtain the second Pending partition value; Use the minimum gray value as the starting point for division, use the second undetermined division value as the division unit, divide a first preset number of gray scale groups, correspond the input set of gray scale values to each gray scale group, and select the gray scale The gray level group with the largest number of distribution values is set as the gray level comparison parameter group; Set the grayscale range of the grayscale comparison parameter group as the grayscale comparison parameter. 5. A method for establishing a thyroid cancer image comparison model based on a deep neural network according to claim 2, characterized in that dividing the thyroid area into an overall area and a point area includes: obtaining a ratio in a two-dimensional coordinate system. For pixels, the area of connected comparison pixels is set as the overall undetermined area; Obtain the number of the overall undetermined area in the two-dimensional coordinate system. When the number of the overall undetermined area is less than or equal to the first distribution quantity threshold, the overall undetermined area is set as the overall area. When the number of the overall undetermined area is greater than the first distribution quantity threshold, , setting the overall undetermined area as a point area. 6. A method for establishing a thyroid cancer image comparison model based on a deep neural network according to claim 5, characterized in that, performing shape feature training on the entire area of several training images, and obtaining the shape comparison parameters includes: setting The outer circle selects the entire area, and the outer circle is the smallest circle that can completely select the entire area; Each time the radius of the outer circle is reduced by the first unit length, an updated circle is obtained, and the center of the updated circle is consistent with the center of the outer circle; The overall area inside the update circle obtained each time is set as the internal area to be divided, and the overall area between the update circle obtained each time and the adjacent update circle or peripheral circle outside is set as the cutting area; Set the area of interconnected comparison pixels in the cutting area as cutting independent areas, count the number of cutting independent areas in each group of cutting areas, and set it as the edge divergence number; when the edge divergence number is less than or equal to the first independent number When the threshold is reached, stop reducing the radius of the update circle or peripheral circle; Obtain the maximum value of the number of edge divergences of several groups and set it as the number of overall divergence distribution; The overall divergence distribution quantity corresponding to several training images is processed through the shape comparison extraction method to obtain the overall comparison parameter, and the overall comparison parameter includes the overall divergence distribution quantity range. 7. A method for establishing a thyroid cancer image comparison model based on a deep neural network according to claim 6, characterized in that, performing shape feature training on point-like areas of several training images, and obtaining shape comparison parameters includes: obtaining The number of point-like areas in the training image is set to the number of point-like distributions; The number of point distributions corresponding to several training images is processed through a shape comparison extraction method to obtain point comparison parameters, where the point comparison parameters include a range of point distribution numbers. 8. A method for establishing a thyroid cancer image comparison model based on a deep neural network according to claim 7, characterized in that the shape comparison extraction method includes: obtaining the maximum value and minimum value of a set of input distribution quantities. The values are respectively set to the maximum value of the distribution amount and the minimum value of the distribution amount; a group of distribution amounts includes a group of the number of overall divergent distributions corresponding to several training images or the number of point distributions corresponding to several training images; Subtract the minimum value of the distribution amount from the maximum value of the distribution amount to obtain the distribution amount difference, divide the distribution amount difference by the second preset number to obtain the first distribution amount division value, extract the integer digit of the first distribution amount division value and add one to obtain The second distribution quantity division value; Use the minimum value of the distribution amount as the starting point for division, use the second distribution amount division value as the division unit, divide the second preset number of distribution amount groups, correspond the input set of distribution amounts to each distribution amount group, and select the distribution The distribution quantity group with the largest quantity distribution is set as the distribution quantity comparison parameter group; The range of the distribution comparison parameter group is set as a shape comparison parameter, and the shape comparison parameter includes an overall comparison parameter and a point comparison parameter.