Submucosal multi-point thickness measuring method for endoscopic submucosal dissection

Abstract

The invention discloses a submucous membrane multi-point thickness measuring method for endoscopic submucous membrane dissection. The submucous membrane multi-point thickness measuring method comprises the following steps that images are collected, wherein pathological images cut by the endoscopic submucous membrane dissection are obtained; image preprocessing: improving the brightness of the pathological image and the contrast ratio of a bright part, and smoothing the pathological image through Gaussian filtering; image feature extraction: extracting texture features of the pathological image, using a local binary pattern (LBP) algorithm, and generating a vertical coordinate feature map with the same size as the pathological image; feature fusion: superposing the preprocessed noise reduction image, the texture feature image and the ordinate feature map; submucous membrane pathological tissue segmentation: using a UNet neural network to carry out image segmentation task training on the superimposed image; calculating the thickness of multiple points; and quality evaluation. According to the method, the thickness of the submucous membrane is measured through the pathological tissue image, and ESD effect judgment and operation quality control are facilitated.

Description

Technical Field

[0001] The present application relates to the technical field of endoscopic submucosal dissection (ESD), and in particular to a method for measuring the multi-point thickness of the submucosal layer used in endoscopic submucosal dissection. Background Art

[0002] Digestive tract diseases are common and frequent health problems worldwide, including but not limited to gastritis, gastric ulcer, colitis, Crohn's disease, gastrointestinal bleeding, and intestinal tumors. These diseases not only have a high incidence rate, but also have complex and diverse symptoms, which have a serious impact on the health and quality of life of patients. With the intensification of the aging trend of the population and changes in lifestyle, the incidence of digestive tract diseases has shown an upward trend, which has seriously affected people's quality of life and health level. Therefore, it is particularly important to accurately diagnose and effectively treat digestive tract diseases.

[0003] Medical endoscopes play an important role in diagnosing gastrointestinal diseases. Traditional endoscopic examinations usually require doctors to insert a long cable into the patient's body to image key areas and provide diagnostic and treatment options. However, this wired endoscope may cause discomfort and pain to the patient due to its invasive nature. In addition, the use of such endoscopes also poses the risk of potential complications, such as infection, perforation, and tearing. Capsule Endoscopy (CE) is an innovative medical device used in the field of modern gastroenterology diagnosis and treatment. It is usually composed of a miniature camera, a housing, a lighting system, a control circuit, a wireless communication module, and a battery. It examines the digestive tract in a painless, non-invasive manner. However, in the process of implementing this application, the applicant found that the current capsule endoscopy detection method has at least the following technical problems: 1. During each test, the wireless capsule endoscope will capture more than 60,000 images. The number of images collected by the capsule endoscope is huge, and manual testing is time-consuming and laborious, which increases the workload of clinicians; 2. Due to problems such as uneven illumination and blurred focus in capsule endoscopy images, feature extraction is difficult. Existing detection methods are difficult to accurately identify irregular lesions under complex backgrounds and lack dedicated detection. 3. Existing models show insufficient feature capture capabilities when detecting tiny lesions, which increases the risk of missed and misdiagnosed lesions, especially in early diagnosis, where critical lesions may be missed. 4. The deep learning models used in the existing technology have insufficient generalization capabilities in specific scenarios and are difficult to cope with various types of gastrointestinal images. In addition, although recent studies have proposed some improvement plans, such as enriching the diversity of data sets through data augmentation technology, or extracting the fractal dimension of images by introducing differential box counting methods to improve detection performance, these methods still fail to effectively solve the problems of insufficient feature extraction and limited detection of small lesions in capsule endoscopy images. In addition, these studies also have deficiencies in data set balance and model generalization capabilities, and cannot fully meet clinical needs.

[0004] Therefore, the above problems lead to low efficiency and low detection accuracy of capsule endoscopy and the risk of missed diagnosis and misdiagnosis, thus limiting the effective application of capsule endoscopy in clinical diagnosis. Summary of the invention

[0005] The embodiment of the present application provides a method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection, which solves the problems of low detection efficiency, low detection accuracy and the risk of missed diagnosis and misdiagnosis of current capsule endoscopes.

[0006] In order to solve the above technical problems, this application is implemented as follows: A method for measuring the thickness of a submucosal layer at multiple points for endoscopic submucosal dissection is provided, which comprises the following steps: Image acquisition: Obtain pathological images cut by endoscopic submucosal dissection; Image preprocessing: improve the brightness and highlight contrast of pathological images, and smooth the pathological images through Gaussian filtering; Image feature extraction: extracting texture features of the pathological image using a local binary pattern (LBP) algorithm, and generating a ordinate feature map of the same size as the pathological image; Feature fusion: superimpose the denoised image, texture feature image and ordinate feature map obtained by preprocessing to form a superimposed image for segmentation; Submucosal pathological tissue segmentation: The UNet neural network is used to train the superimposed images for image segmentation tasks to extract the submucosal layer; Multi-point thickness calculation: The skeleton of the submucosal layer is extracted using the skeletonization method, and the skeleton is used as the extension direction of the submucosal layer at that position. The vertical straight line is used to calculate the thickness of the submucosal layer at that position, so as to achieve the multi-point thickness measurement of the submucosal layer during endoscopic submucosal dissection. Quality assessment: The image was divided into equal parts vertically by manually selecting the endpoints of the straight line and performing an equal division operation to assess the surgical quality.

[0007] In one embodiment, in the step of improving the brightness of the pathological image: Adjust the background brightness of the pathological image, adjust the brightness of the V channel of the image in the HSV color space, and retain the hue (H) and saturation (S) channels to avoid image color distortion. The clipping histogram is defined as follows: , represents the original histogram, , clipping threshold Defined as 180.

[0008] In one embodiment, in the step of enhancing the contrast of the bright part of the pathological image: Gamma transform is used to perform nonlinear transformation on the image pixel values ​​to improve the contrast of the bright part of the pathological image. The formula of gamma transform is: , A is the transformation coefficient, which is used to restore the transformed image to the original image bit space, γ is used as the exponent of the exponential calculation, and when γ is greater than 1, the image contrast in the high brightness area can be improved.

[0009] In one embodiment, in the step of smoothing the pathological image by Gaussian filtering: The pathological image is processed by applying Gaussian filtering, using a standard deviation value to smooth the image and reduce the interference of noise and details. The Gaussian filtering formula is: , G(x,y) is the image value after Gaussian filtering, x and y are the image pixel coordinates, and σ is the standard deviation of the Gaussian distribution.

[0010] In one embodiment, in the step of extracting texture features of a pathological image: Convert color images of pathological images into grayscale images; The LBP algorithm is used to process grayscale images. The LBP algorithm generates a texture feature map by comparing the grayscale values ​​of each pixel with its neighboring pixels.

[0011] In one embodiment, the model architecture of the UNet neural network includes: Convolution module: includes two convolution operations, and batch normalization and ReLU activation function processing are performed after each convolution; Downsampling path: reduce the feature map resolution and increase the number of channels through convolution modules and maximum pooling operations; Bottleneck layer: The features are further processed and compressed through the convolution module to obtain the most representative high-dimensional feature representation; Feature splicing: The UNet model splices the features in the downsampling process to the upsampled features according to the feature level for subsequent feature upsampling processing; Upsampling path: During the upsampling process, the deconvolution operation is used to gradually restore the resolution of the feature map. At the same time, the upsampled features are fused with the corresponding downsampling layer features through feature concatenation, and then the features are integrated and optimized through the convolution module. Output layer: A convolutional layer is used to map the features to the number of output channels. This method is a two-class image segmentation with an output channel of 1. The Sigmoid activation function is used to limit the output value between 0 and 1 to obtain the probability that each pixel belongs to the target category.

[0012] In one embodiment, the model architecture of the UNet neural network also includes: Position attention mechanism: used to encode and weight the position information of the feature map before feature fusion or jump connection, so as to improve the model's sensitivity to position features and improve the model's segmentation accuracy.

[0013] In one embodiment, in the step of weighting the position information of the feature map: Perform a linear transformation on the input feature map to obtain the query vector Q, key vector K and value vector V; Calculate the dot product of Q and K through matrix multiplication, and then perform scaling and Softmax operations to obtain the attention weight matrix A; Multiply the attention weight matrix A with the value vector V to obtain the weighted feature representation to enhance the model's ability to recognize the region of interest.

[0014] In one embodiment, in the step of using a UNet neural network to train the superimposed image for an image segmentation task: Model training: The error loss is calculated by combining the Dice parameter with the mean square error, and the model parameters are optimized using the stochastic gradient descent algorithm to adapt to image segmentation under the condition of 5-channel image feature input; Processing segmented images: For false positive points and false negative points in the segmentation results, they are corrected by connecting area filtering, the holes of false negative points are closed by hole filling combined with closing operations, and the edges of false negative points are processed by edge smoothing operations to ensure the accuracy of the segmentation results.

[0015] In one embodiment, in the step of closing the holes of false negative points of the segmented image using hole filling combined with a closing operation: Determine the hole area: traverse each negative connected area in the segmented image, and for each negative connected area, check the pixels in the outer 8 connected areas. If all pixels are positive, mark the negative connected area as a hole area; Hole filling: Modify all negative pixel values ​​in each marked hole area to positive pixel values.

[0016] In the embodiment of the present application, the pathological image of the surgical dissection tissue is preprocessed, and then a variety of feature extraction methods are used for feature enhancement, and finally a convolutional neural network is used to segment the submucosal layer. At the same time, closed operations and connected area filtering are used to remove erroneous results. Combined with the backbone direction positioning method, the multi-point thickness measurement of the submucosal layer of the pathological tissue is finally achieved. In this way, the present application measures the thickness of the submucosal layer through pathological tissue images, which is helpful for the effect determination and surgical quality control of endoscopic submucosal dissection (ESD). BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 is a flowchart of the steps of the method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection of the present application; Figure 2 It is the gamma transformation function diagram when γ is 5 in this application; Figure 3 and Figure 4 is a schematic diagram of a gastric wall submucosal specimen image before and after gamma transformation of the present application; Figure 5 and Figure 6 is a schematic diagram of a gastric wall submucosal specimen image before and after Gaussian filtering processing of the present application; Figure 7 It is a schematic diagram of the LBP feature extraction effect of this application; Figure 8 It is the image longitudinal position information feature map of the present application; Fig. 9 is a schematic diagram of the superimposed images of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] See also Figure 1, which is a flowchart of the steps of the multi-point thickness measurement method of the submucosal layer for endoscopic submucosal dissection of the present application; as shown in the figure, the multi-point thickness measurement method of the submucosal layer for endoscopic submucosal dissection includes the following steps S1 to S7. Among them: Step S1, image acquisition: obtaining pathological images cut by endoscopic submucosal dissection. Pathological images are obtained from the cut pathological tissue obtained by endoscopic submucosal dissection, provided by hospital experts, and uploaded to computers and storage devices. The collected images are processed to obtain a series of images. , the pathological sample data is expressed as .

[0020] Step S2, image preprocessing: enhance the brightness and highlight contrast of the pathological image, and smooth the pathological image through Gaussian filtering. The pathological images collected in the experiment may have uneven brightness, unclear edge information, etc. The first step is to use color and brightness screening to enhance the brightness of the image background, and set the image background to the highest brightness according to the color brightness threshold. Specifically, since the shooting environment of the pathological tissue may have poor lighting conditions, first adjust the background brightness of the pathological image, adjust the brightness of the V channel of the image in the HSV color space, and retain the hue (H) and saturation (S) channels to avoid image color distortion. This method can avoid over-enhancing the brightness of the image, and the cropped histogram is defined as follows: , represents the original histogram, , clipping threshold Defined as 180.

[0021] As mentioned above, the second step is to correct the image brightness and improve the contrast of the bright part based on the gamma correction method. Specifically, since the image brightness of the pathological tissue and the background area in the gastric wall submucosal specimen image is high, its histogram is distributed in the high brightness area, and the contrast between the pathological tissue and the background is poor, the gamma transform is used to perform nonlinear transformation on the image pixel value to improve the contrast of the bright part of the pathological image. The formula of the gamma transform is The principle is to perform exponential calculation on the input image to achieve nonlinear transformation. Among them, A is the transformation coefficient, which is used to restore the transformed image to the original image bit space. γ is the exponent of the exponential calculation. When γ is greater than 1, the image contrast in the high-brightness area can be improved. The image is processed by applying gamma transformation, and the gamma value is selected as 5 to enhance the contrast of the bright area. The transformation function image is as follows Figure 2This processing method effectively improves the detail performance of the highlight area in the submucosal specimen image, making the important information more prominent. Since the shooting scene of the gastric submucosal specimen image is bright, the gamma transform method is used to enhance the contrast of the bright part. The effect is shown in Figure 3 and Figure 4 shown.

[0022] The third step is to remove image noise through Gaussian filtering and remove redundant features of the image. By applying Gaussian filtering to process the pathological image, a standard deviation value σ is used to smooth the image to reduce the interference of noise and details. This processing method can effectively suppress random noise in the submucosal specimen image while maintaining the integrity of important structures such as edges. The Gaussian filter formula is: , G(x,y) is the image value after Gaussian filtering, x and y are the image pixel coordinates, and σ is the standard deviation of the Gaussian distribution.

[0023] Figure (4) Gaussian filter sliding window By selecting the appropriate σ value, Gaussian filtering can effectively retain key edge information when smoothing the image, making the features of the submucosal specimen image clearer. In this digital image processing task, a convolution kernel of size 3×3 is used to perform Gaussian filtering on the discrete digital image to reduce noise while retaining key details. The effect is as follows Figure 5 and Figure 6 As shown in Figure 2, Gaussian filtering removes the noise of the image to a certain extent and optimizes the holes and noise points in the pathological image of the gastric wall submucosal specimen.

[0024] Step S3, image feature extraction: extract the texture features of the pathological image, use the local binary pattern (LBP) algorithm, and generate a ordinate feature map of the same size as the pathological image. This embodiment uses an efficient image texture feature extraction method, which first converts the input color image into a grayscale image. In other words, the color image of the pathological image is converted into a grayscale image, so as to reduce the complexity of the calculation and highlight the brightness information of the image. Subsequently, the local binary pattern (LBP) algorithm is used to process the grayscale image. The LBP algorithm generates a texture feature map by comparing the grayscale values ​​of each pixel with its neighboring pixels. The specific principle is to compare the brightness of the pixel point with the surrounding pixel points within a certain range. If it is higher than the comparison point, the feature is recorded as 1, otherwise it is 0, and a decimal value is generated based on this binary sequence, which is used as the texture feature of the point. In this embodiment, the number of neighborhood pixels is set to 8, the neighborhood radius is set to 5, and the decimal number is simplified when it is generated. The equivalent mode is used to only calculate the number of changes between 0 and 1, so as to simplify the complexity of the feature and obtain a more robust and discriminative feature description. The LBP feature extraction effect is as follows: Figure 7 shown.

[0025] As mentioned above, we will then define the image longitudinal position information features. Since we need to calculate the thickness of the submucosal layer in the pathological image of the gastric wall submucosal specimen, we need to segment the image of this part. Since the submucosal tissue has a relatively obvious position feature and is located at the bottom of the case tissue, we define the image longitudinal position information features, that is, generate a longitudinal coordinate feature map with the same size as the image, and the effect is as follows: Figure 8 shown.

[0026] Step S4, feature fusion: superimpose the denoised image, texture feature image and ordinate feature map obtained by preprocessing to form a superimposed image for segmentation. This method superimposes the images obtained from steps S1 to S3 as input data for image segmentation, and the effect is as follows: Fig. 9 As shown, the input is given to the following UNet neural network.

[0027] Step S5, submucosal pathological tissue segmentation: Use the UNet neural network to train the superimposed image for image segmentation tasks to extract the submucosal layer. UNet is a deep learning model architecture commonly used for image segmentation tasks. The model architecture of the UNet neural network includes: Convolution module: The convolution module contains two convolution operations, and batch normalization and ReLU activation function processing are performed after each convolution.

[0028] Downsampling path: Reduce the resolution of the feature map and increase the number of channels through convolution modules and maximum pooling operations. In the downsampling process, the resolution of the feature map is gradually reduced through a series of convolution modules and maximum pooling operations, while the number of channels is increased, thereby extracting high-level abstract features of the image. Specifically, starting from the number of input channels, after multiple downsampling, the number of channels gradually increases and the size of the feature map gradually decreases.

[0029] Bottleneck layer: The features are further processed and compressed through the convolution module to obtain the most representative high-dimensional feature representation.

[0030] Feature splicing: The UNet model splices the features from the downsampling process onto the upsampled features according to the feature level for subsequent feature upsampling. The purpose of this is to fuse the coarser but semantically rich features obtained from upsampling with the finer but semantically shallow features preserved during the downsampling process, thereby comprehensively utilizing feature information at different levels to improve the segmentation performance of the model. In this way, the model can better capture the global structure and local details of the image, thereby achieving more accurate segmentation results.

[0031] Upsampling path: During the upsampling process, the deconvolution operation is used to gradually restore the resolution of the feature map. At the same time, the upsampled features are fused with the corresponding downsampling layer features through feature splicing, and then the features are integrated and optimized through the convolution module.

[0032] Output layer: A convolutional layer is used to map the features to the number of output channels. This method is a two-class image segmentation with an output channel of 1. The Sigmoid activation function is used to limit the output value between 0 and 1 to obtain the probability that each pixel belongs to the target category.

[0033] In this embodiment, the model architecture of the UNet neural network also includes a position attention mechanism. Since the submucosal layer has significant positional features in the image, that is, it appears in the middle and lower parts of the image, in order to enhance the model with positional attention, it is used to improve the sensitivity of the model to positional features and improve the segmentation accuracy of the model. Positional attention mechanism: used to encode and weight the position information of the feature map before feature fusion or jump connection, so as to improve the sensitivity of the model to positional features and improve the segmentation accuracy of the model. In the step of weighting the position information of the feature map: first, the input feature map is linearly transformed to obtain the query vector Q, the key vector K and the value vector V; then, the dot product of Q and K is calculated by matrix multiplication, and then scaling and Softmax operations are performed to obtain the attention weight matrix A; finally, the attention weight matrix A is multiplied by the value vector V to obtain the weighted feature representation to enhance the model's recognition ability of the region of interest.

[0034] In the steps of training the image segmentation task of superimposed images using the UNet neural network: Model training: and use the stochastic gradient descent algorithm to optimize the model parameters to adapt to image segmentation under the working conditions of 5-channel image feature input.

[0035] As mentioned above, the segmented image is processed: for the false positive points and false negative points in the segmentation results, the connected area filtering method is used to correct them, the holes of the false negative points are closed by hole filling combined with closing operation, and the edges of the false negative points are processed by edge smoothing operation to ensure the accuracy of the segmentation results. False positive points usually appear as erroneous discrete points. In order to correct them, the connected area filtering method is used to correct the false positive points. The connected area refers to the area composed of adjacent pixels with the same attribute value in the image. I(x,y) represents the segmentation result image, (x,y) represents the coordinates of the pixel point, and for each connected area, its area A can be calculated. The area with the largest area among all connected areas is recorded as A{max}. When correcting false positive points, if there is a connected area with an area Ai smaller than the area A{max} of the largest connected area and is marked as positive, the corresponding connected area will be removed from the positive result.

[0036] In this way, the false positive results are corrected, thereby improving the accuracy of the segmentation results.

[0037] The existence of false negative points in the segmentation results is in the form of holes and missing edges in the segmentation results. Holes of false negative points in the segmented image are first closed by hole filling combined with closing operations, and then edge smoothing operations are used to handle false negative errors at the edges of false negative points in the segmented image. The hole filling algorithm: 1. Determine the hole area: traverse each negative connected area in the segmented image, and for each negative connected area, check the pixels in the 8 connected areas outside it. If all pixels are positive, mark the negative connected area as a hole area; 2. Hole filling: modify all negative pixel values ​​of each marked hole area to positive pixel values. When determining the hole area, the specific method of checking the 8-connected area is: for the edge pixel (x, y) of the current negative connected area, the pixel coordinates of its 8-connected area are (x - 1, y - 1), (x - 1, y), (x - 1, y + 1), (x, y - 1), (x, y + 1), (x +1, y - 1), (x + 1, y), (x + 1, y + 1). Check the pixel values ​​of these positions in turn to see if the pixels that do not belong to the connected area are all positive.

[0038] Step S6, multi-point thickness calculation: use the skeletonization method to extract the submucosal skeleton, use the skeleton as the extension direction of the submucosal layer at this position, and use the vertical straight line to calculate the thickness of the submucosal layer at this position, so as to achieve the multi-point thickness measurement of the submucosal layer in endoscopic submucosal dissection. Since the image segmented in step S5 is the submucosal tissue in the pathological image, the multi-point thickness measurement method of the submucosal layer is to measure the thickness of the segmentation result. Since the submucosal layer in the pathological image does not strictly meet the horizontal distribution, it is necessary to determine the correct thickness of the submucosal layer and the extension direction of the submucosal tissue.

[0039] Step S7, quality assessment: manually select the endpoints of the straight line and perform equal division operations to divide the image into equal parts in the vertical direction to evaluate the quality of the surgery. The manually defined positions are used as the endpoints of the straight line, and four or five equal division points are located. Several straight lines perpendicular to the straight line at these points are used to divide the image into equal parts. When the manually selected positioning point is outside the segmentation result, the straight line is removed by up to 5% in this direction to remove tissue outside the segmentation result on the straight line.

[0040] In summary, the present application provides a method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection, which pre-processes the pathological images of the surgical dissection tissue, and then uses a variety of feature extraction methods to enhance the features, and finally uses a convolutional neural network to segment the submucosal layer. At the same time, the closed operation and connected area filtering methods are used to remove erroneous results, combined with the backbone direction positioning method, and finally realizes the multi-point thickness measurement of the submucosal layer of the pathological tissue. In this way, the present application measures the thickness of the submucosal layer through pathological tissue images, which is helpful for the effect judgment and surgical quality control of endoscopic submucosal dissection (ESD).

[0041] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0042] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection, characterized in that: The following steps are involved: Image acquisition: Obtain pathological images cut by endoscopic submucosal dissection; Image preprocessing: improving the brightness and highlight contrast of the pathological image, and smoothing the pathological image by Gaussian filtering; Image feature extraction: extracting texture features of the pathological image using a local binary pattern (LBP) algorithm, and generating a ordinate feature map having the same size as the pathological image; Feature fusion: superimpose the denoised image, texture feature image and ordinate feature map obtained by preprocessing to form a superimposed image for segmentation; Segmentation of submucosal pathological tissue: using a UNet neural network to train the superimposed image for an image segmentation task to extract the submucosal layer; Multi-point thickness calculation: The skeleton of the submucosal layer is extracted using the skeletonization method, and the skeleton is used as the extension direction of the submucosal layer at that position. The vertical straight line is used to calculate the thickness of the submucosal layer at that position, so as to achieve the multi-point thickness measurement of the submucosal layer during endoscopic submucosal dissection. Quality assessment: The image was divided into equal parts vertically by manually selecting the endpoints of the straight line and performing an equal division operation to assess the surgical quality.

2. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 1, characterized in that: In the step of improving the brightness of the pathological image: The background brightness of the pathological image is adjusted, the brightness of the V channel of the image is adjusted in the HSV color space, and the hue (H) and saturation (S) channels are retained to avoid image color distortion, wherein the clipping histogram is defined as follows: , represents the original histogram, , clipping threshold Defined as 180.

3. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 1, characterized in that: In the step of improving the bright part contrast of the pathological image: Gamma transform is used to perform nonlinear transformation on the image pixel values ​​to improve the contrast of the bright part of the pathological image. The formula of gamma transform is: , A is the transformation coefficient, which is used to restore the transformed image to the original image bit space, γ is used as the exponent of the exponential calculation, and when γ is greater than 1, the image contrast in the high brightness area can be improved.

4. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 1, characterized in that: In the step of smoothing the pathological image by Gaussian filtering: The pathological image is processed by applying Gaussian filtering, using a standard deviation value to smooth the image and reduce the interference of noise and details, wherein the Gaussian filtering formula is: , G(x,y) is the image value after Gaussian filtering, x and y are the image pixel coordinates, and σ is the standard deviation of the Gaussian distribution.

5. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 1, characterized in that: In the step of extracting texture features of the pathological image: converting the color image of the pathological image into a grayscale image; The LBP algorithm is used to process the grayscale image. The LBP algorithm generates the texture feature map by comparing the grayscale values ​​of each pixel with its neighboring pixels.

6. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 1, characterized in that: The model architecture of the UNet neural network includes: Convolution module: includes two convolution operations, and batch normalization and ReLU activation function processing are performed after each convolution; Downsampling path: reduce the feature map resolution and increase the number of channels through convolution modules and maximum pooling operations; Bottleneck layer: The features are further processed and compressed through the convolution module to obtain the most representative high-dimensional feature representation; Feature splicing: The UNet model splices the features in the downsampling process to the upsampled features according to the feature level for subsequent feature upsampling processing; Upsampling path: During the upsampling process, the deconvolution operation is used to gradually restore the resolution of the feature map. At the same time, the upsampled features are fused with the corresponding downsampling layer features through feature concatenation, and then the features are integrated and optimized through the convolution module. Output layer: A convolutional layer is used to map the features to the number of output channels. This method is a two-class image segmentation with an output channel of 1. The Sigmoid activation function is used to limit the output value between 0 and 1 to obtain the probability that each pixel belongs to the target category.

7. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 6, characterized in that: The model architecture of the UNet neural network also includes: Position attention mechanism: used to encode and weight the position information of the feature map before feature fusion or jump connection, so as to improve the model's sensitivity to position features and improve the model's segmentation accuracy.

8. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 7, characterized in that: In the step of weighting the location information of the feature map: Perform a linear transformation on the input feature map to obtain the query vector Q, key vector K and value vector V; Calculate the dot product of Q and K through matrix multiplication, and then perform scaling and Softmax operations to obtain the attention weight matrix A; Multiply the attention weight matrix A with the value vector V to obtain the weighted feature representation to enhance the model's ability to recognize the region of interest.

9. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 1, characterized in that: In the step of using the UNet neural network to train the superimposed image for an image segmentation task: Model training: The error loss is calculated by combining the Dice parameter with the mean square error, and the model parameters are optimized using the stochastic gradient descent algorithm to adapt to image segmentation under the condition of 5-channel image feature input; Processing segmented images: For false positive points and false negative points in the segmentation results, they are corrected by connecting area filtering, the holes of false negative points are closed by hole filling combined with closing operations, and the edges of false negative points are processed by edge smoothing operations to ensure the accuracy of the segmentation results.

10. The method for measuring the multi-point thickness of the submucosal layer for endoscopic submucosal dissection according to claim 9, characterized in that: In the step of closing the holes of false negative points of the segmented image using hole filling combined with closing operation: Determine the hole area: traverse each negative connected area in the segmented image, and for each negative connected area, check the pixels in the outer 8 connected areas. If all pixels are positive, mark the negative connected area as a hole area; Hole filling: Modify all negative pixel values ​​in each marked hole area to positive pixel values.

Images