Enteroscope image segmentation method and device, equipment, storage medium and computer program product

By calculating the similarity and confidence of pixel points in the colonoscopic image, and using image annotation and segmentation models, the low accuracy caused by noise in intestinal ulcer segmentation technology is solved, and the accuracy and diagnostic effect of image segmentation are improved.

CN118968048APending Publication Date: 2024-11-15SHENZHEN UNIV
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202410937286.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

When marking colonoscopic images, the existing intestinal ulcer segmentation technology is prone to noise due to the doctor's judgment and manual smearing limitations, resulting in low image segmentation accuracy and affecting the accuracy of diagnosis.

Method used

By determining the similarity between pixel points and adjacent pixel points in the colonoscopic image, the confidence of pixel points is calculated based on the similarity, and the confidence annotation model is used to mark the colonoscopic image, and then the preset image segmentation model is performed. Segmentation to reduce noise impact.

Benefits of technology

It improves the accuracy of colonoscopic image segmentation, enhances the accuracy of diagnosis, and reduces the impact of noise on segmentation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968048A_ABST
    Figure CN118968048A_ABST
Patent Text Reader

Abstract

The invention discloses an enteroscopy image segmentation method, device and equipment, a storage medium and a computer program product, relates to the technical field of image processing, and discloses an enteroscopy image segmentation method, which comprises the following steps: determining pixel point similarity between all pixel points in a to-be-segmented enteroscopy image and corresponding adjacent pixel points; determining pixel point confidence corresponding to each pixel point based on the pixel point similarity; performing confidence degree labeling on pixel point labels in the to-be-segmented enteroscopy image based on the pixel point confidence degree through a preset image labeling model to obtain a labeled enteroscopy image; and segmenting the to-be-segmented enteroscope image based on the labeled enteroscope image through a preset image segmentation model to obtain an enteroscope segmented image. According to the technical scheme, the technical problem that in the prior art, noise is likely to occur in the process of labeling the enteroscopy image through the intestinal ulcer segmentation technology, and consequently the image segmentation precision is low is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method, device, equipment, storage medium and computer program product for colonoscopy image segmentation. Background Art

[0002] Inflammatory bowel disease (IBD) is a chronic nonspecific intestinal inflammatory disease, the exact cause of which is still not fully understood. With its increasing incidence rate, IBD has become the focus of global medical attention. However, although the advancement of diagnostic technology has enabled more and more children to be diagnosed with IBD, the harm to children's health cannot be ignored due to the complex etiology and relatively simple treatment methods.

[0003] Currently, intestinal ulcer segmentation technology can be used to accurately segment colonoscopy images to assist doctors in observation, thus helping to diagnose inflammatory bowel disease. However, in the process of annotating colonoscopy images, due to the limitations of doctors' judgment and manual painting, noise will inevitably appear in the labeled images, resulting in low accuracy of subsequent image segmentation, which in turn affects the accuracy of diagnosis.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present application is to provide a colonoscopy image segmentation method, device, equipment, storage medium and computer program product, aiming to solve the technical problem that in the prior art, intestinal ulcer segmentation technology is prone to noise during the annotation of colonoscopy images, resulting in low image segmentation accuracy.

[0006] To achieve the above objectives, the present application proposes a colonoscopy image segmentation method, the method comprising:

[0007] Determine the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels;

[0008] Determine the pixel confidence corresponding to each pixel based on the pixel similarity;

[0009] Performing confidence annotation on pixel labels in the colonoscopy image to be segmented based on the pixel confidence using a preset image annotation model to obtain an annotated colonoscopy image;

[0010] The colonoscopy image to be segmented is segmented based on the annotated colonoscopy image using a preset image segmentation model to obtain a colonoscopy segmentation image.

[0011] In one embodiment, before the step of determining the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels, the method further includes:

[0012] Inputting a plurality of colonoscopy training images into a preset image annotation model, so as to annotate each colonoscopy training image through the preset image annotation model, and obtaining a plurality of annotated colonoscopy training images, wherein the annotated colonoscopy training images are annotated with corresponding image labels and confidence masks, and the confidence masks are used to identify noise areas in the image labels;

[0013] constructing a semantic segmentation dataset based on the annotated colonoscopy training image, the image label and the confidence mask;

[0014] The collaboratively trained segmentation model is trained based on the semantic segmentation dataset to obtain a preset image segmentation model.

[0015] In one embodiment, the step of inputting a plurality of colonoscopy training images into a preset image annotation model, annotating each colonoscopy training image by the preset image annotation model, and obtaining a plurality of annotated colonoscopy training images includes:

[0016] Input a number of colonoscopy training images into a preset image annotation model;

[0017] Determine the pixel distance between the target pixel and the corresponding target adjacent pixel in each colonoscopy training image by using the preset image annotation model;

[0018] Determine a distance weight corresponding to the target pixel point based on the pixel point distance;

[0019] Determine the similarity between the target pixel and the target adjacent pixel based on the distance weight;

[0020] Each colonoscopy training image is annotated based on the similarity using the preset image annotation model to obtain a plurality of annotated colonoscopy training images.

[0021] In one embodiment, the step of annotating each colonoscopy training image based on the similarity by using the preset image annotation model to obtain a plurality of annotated colonoscopy training images includes:

[0022] Determining the confidence of the target pixel based on the similarity;

[0023] Determining a confidence mask corresponding to the target pixel according to the confidence level;

[0024] The image annotation model is used to annotate each colonoscopy training image based on the similarity and the confidence mask to obtain a plurality of annotated colonoscopy training images.

[0025] In one embodiment, the collaboratively trained segmentation model is composed of a first image segmentation network and a second image segmentation network; the first image segmentation network and the second image segmentation network both adopt the Unet++ architecture, and the Unet++ architecture is designed with a skip path and deep supervision; the skip path is used to reduce the semantic gap between the feature maps of the encoder and decoder subnets, and the deep supervision is used to segment lesions.

[0026] In one embodiment, the step of performing model training on the collaboratively trained segmentation model based on the semantic segmentation dataset to obtain a preset image segmentation model includes:

[0027] Performing model training on the collaborative training segmentation model based on the semantic segmentation dataset to obtain an initial image segmentation model;

[0028] Determine the uncertainty score corresponding to the sample colonoscopy image and the diversity score corresponding to the unlabeled sample colonoscopy image;

[0029] Determine a target sample colonoscopy image based on the uncertainty score and the diversity score;

[0030] Annotating the target sample colonoscopy image based on the expert annotation information to obtain an annotated sample colonoscopy image;

[0031] Adding the labeled sample colonoscopy images to the original source data set to obtain an updated original source data set;

[0032] The initial image segmentation model is iteratively trained based on the updated original source data set to obtain a preset image segmentation model.

[0033] In addition, to achieve the above-mentioned purpose, the present application also proposes a colonoscopy image segmentation device, the device comprising:

[0034] A similarity determination module is used to determine the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels;

[0035] A confidence determination module, used to determine the pixel confidence corresponding to each pixel based on the pixel similarity;

[0036] An image annotation module, used to perform confidence annotation on pixel labels in the colonoscopy image to be segmented based on the pixel confidence using a preset image annotation model to obtain an annotated colonoscopy image;

[0037] An image segmentation module is used to segment the colonoscopic image to be segmented based on the annotated colonoscopic image using a preset image segmentation model to obtain a colonoscopic segmented image.

[0038] In addition, to achieve the above-mentioned purpose, the present application also proposes a colonoscopy image segmentation device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the colonoscopy image segmentation method described above.

[0039] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the colonoscopy image segmentation method described above are implemented.

[0040] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the colonoscopy image segmentation method described above.

[0041] The present application provides a colonoscopy image segmentation method, which discloses determining the pixel similarity between all pixels in a colonoscopy image to be segmented and corresponding adjacent pixels; determining the pixel confidence corresponding to each pixel based on the pixel similarity; annotating the pixel labels in the colonoscopy image to be segmented based on the pixel confidence by a preset image annotation model to obtain the annotated colonoscopy image; segmenting the colonoscopy image to be segmented based on the annotated colonoscopy image by a preset image segmentation model to obtain the colonoscopy segmentation image; compared with the intestinal ulcer segmentation technology in the prior art, in the process of annotating colonoscopy images, noise appears in the label image due to the doctor's judgment and the limitations of manual painting, resulting in low image segmentation accuracy, the present invention uses a preset image annotation model to annotate the pixel labels in the image based on the pixel confidence of each pixel in the colonoscopy image to be segmented, obtains the annotated colonoscopy image, and performs image segmentation based on the annotated colonoscopy image by a preset image segmentation model, thereby solving the technical problem that the intestinal ulcer segmentation technology in the prior art is prone to noise in the process of annotating colonoscopy images, resulting in low image segmentation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 A schematic diagram of a flow chart provided for the first embodiment of the colonoscopy image segmentation method of the present application;

[0045] Figure 2 A flow chart of the second embodiment of the colonoscopy image segmentation method of the present application;

[0046] Figure 3 This is a diagram of the architecture of the image segmentation model in the colonoscopy image segmentation method of this application;

[0047] Figure 4 This is a flow chart of the confidence region extraction module in the colonoscopy image segmentation method of this application;

[0048] Figure 5 This is a diagram of the architecture of the collaborative training segmentation model in the colonoscopy image segmentation method of this application;

[0049] Figure 6 A flow chart of the third embodiment of the colonoscopy image segmentation method of the present application;

[0050] Figure 7 A training flow chart of a preset image annotation model in the colonoscopy image segmentation method of this application;

[0051] Figure 8 This is a schematic diagram of the module structure of the colonoscopy image segmentation device according to an embodiment of the present application;

[0052] Fig. 9 Schematic diagram of the device structure of the hardware operating environment involved in the colonoscopy image segmentation method in the embodiment of the present application.

[0053] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0054] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0055] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0056] The main solution of the embodiment of the present application is: determine the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels; determine the pixel confidence corresponding to each pixel based on the pixel similarity; perform confidence annotation on the pixel labels in the colonoscopy image to be segmented based on the pixel confidence through a preset image annotation model to obtain a labeled colonoscopy image; perform segmentation on the colonoscopy image to be segmented based on the labeled colonoscopy image through a preset image segmentation model to obtain a colonoscopy segmented image.

[0057] In the process of annotating colonoscopy images, the existing intestinal ulcer segmentation technology causes noise in the label image due to the limitations of doctors' judgment and manual painting, resulting in low image segmentation accuracy.

[0058] The present application provides a solution, which uses a preset image annotation model to perform confidence annotation on pixel labels in an image based on the pixel confidence of each pixel in the colonoscopy image to be segmented, obtains a labeled colonoscopy image, and uses a preset image segmentation model to perform image segmentation based on the labeled colonoscopy image, thereby solving the technical problem in the prior art that intestinal ulcer segmentation technology is prone to noise during the annotation of colonoscopy images, resulting in low image segmentation accuracy.

[0059] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a colonoscopy image segmentation device, etc. The following takes the colonoscopy image segmentation device as an example (hereinafter referred to as the device) to illustrate this embodiment and the following embodiments.

[0060] Based on this, the present application embodiment provides a colonoscopy image segmentation method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the colonoscopy image segmentation method of the present application.

[0061] In this embodiment, the colonoscopy image segmentation method includes steps S10 to S40:

[0062] Step S10: Determine the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels.

[0063] It should be noted that the above-mentioned colonoscopic image to be segmented can be any image collected when performing a colonoscopic examination on a user.

[0064] It should be understood that the adjacent pixels may be pixels that are closer to each pixel in the colonoscopy image to be segmented. Correspondingly, the pixel similarity may be the similarity between each pixel and its corresponding adjacent pixel. In practical applications, the more similar the color and texture of a pixel in the colonoscopy image to be segmented are to its surrounding adjacent pixels, the higher the similarity between the pixel and its adjacent pixels.

[0065] Step S20: determining the pixel confidence corresponding to each pixel based on the pixel similarity.

[0066] It should be noted that the above-mentioned pixel confidence can be a value used to evaluate the credibility of the model's classification or prediction results for each pixel. In practical applications, if the color, texture and other feature similarities between the pixel in the colonoscopy image to be segmented and the surrounding pixels are higher, then the confidence of the application in marking the pixel will be higher. Therefore, this solution can calculate the confidence of the pixel based on the similarity between the pixel and the surrounding pixels; conversely, if the pixel in the colonoscopy image to be segmented has a large feature difference from the surrounding pixels, then the confidence of the application in marking the pixel will be lower. Among them, the model will pay more attention to pixels with higher confidence, thereby improving the prediction accuracy and robustness of the model.

[0067] Step S30: Confidence annotation of pixel labels in the colonoscopy image to be segmented is performed based on the pixel confidence using a preset image annotation model to obtain an annotated colonoscopy image.

[0068] It should be noted that the above-mentioned preset image annotation model can be a model for adding labels to pixel points in the segmented colonoscopy image, so that the subsequent preset image segmentation model can perform image segmentation based on the image annotated by the preset image annotation model, thereby improving the accuracy of image segmentation and further improving the accuracy of disease diagnosis.

[0069] It is understandable that the annotated colonoscopy image may be an image annotated with labels and confidences corresponding to each pixel. In practical applications, after the preset image annotation model calculates and obtains the confidence corresponding to each pixel in the colonoscopy image to be segmented, it can annotate it according to the confidence corresponding to each pixel, thereby obtaining an annotated colonoscopy image annotated with the confidence corresponding to each pixel.

[0070] In this embodiment, if the confidence of a pixel is greater than or equal to a preset similarity threshold, the pixel can be regarded as a point with higher confidence; if the confidence of a pixel is less than the preset similarity threshold, the pixel can be regarded as a pixel with edge noise. The device can then better screen out pixels with higher confidence by setting appropriate thresholds, while excluding edge pixels that may contain noise, thereby helping to improve the accuracy and stability of the segmentation results, and better cope with image data with complex backgrounds and noise.

[0071] Step S40: segmenting the colonoscopic image to be segmented based on the annotated colonoscopic image using a preset image segmentation model to obtain a colonoscopic segmented image.

[0072] It should be noted that the above-mentioned preset image segmentation model may be a model for segmenting the ulcer area in the colonoscopy image to be segmented.

[0073] In this embodiment, the higher the similarity between a pixel point in the colonoscopy image to be segmented and its surrounding adjacent pixel points, the lower the possibility that the pixel point is located at the marked edge, and the correspondingly higher its confidence level, so that the pixel points located at the edge of the ulcer in the colonoscopy image to be segmented can be identified based on the confidence levels of all pixels in the annotated colonoscopy image, so that the preset image segmentation model can segment the ulcer area in the colonoscopy image to be segmented according to the annotated colonoscopy image to obtain a colonoscopy segmented image.

[0074] In practical applications, due to the limitations of doctors’ judgment and manual painting, noise is inevitable in the labeled images. This noise may appear in the edge areas of the image. At this time, if the model is directly trained using these noisy labels, it is likely to be biased in prediction. Therefore, this solution can conduct an in-depth analysis of the confidence information in the labeled data, so that high-confidence areas can be accurately identified and extracted. This step is crucial for training high-quality segmentation models. It ensures that the model can focus on the most reliable data areas during training, which often have the largest amount of information. This can significantly improve the training efficiency of the segmentation model in the future, while improving the accuracy and stability of the segmentation results.

[0075] The present embodiment provides a colonoscopy image segmentation method, which discloses determining the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels; determining the pixel confidence corresponding to each pixel based on the pixel similarity; annotating the pixel labels in the colonoscopy image to be segmented based on the pixel confidence through a preset image annotation model to obtain a labeled colonoscopy image; segmenting the colonoscopy image to be segmented based on the annotated colonoscopy image through a preset image segmentation model to obtain a colonoscopy segmentation image; compared with the intestinal ulcer segmentation technology in the prior art, in the process of annotating colonoscopy images, noise appears in the label image due to the doctor's judgment and the limitations of manual painting, resulting in low image segmentation accuracy, because the present embodiment uses a preset image annotation model to annotate the pixel labels in the image based on the pixel confidence of each pixel in the colonoscopy image to be segmented to obtain a labeled colonoscopy image, and performs image segmentation based on the annotated colonoscopy image through the preset image segmentation model, thereby solving the technical problem that the intestinal ulcer segmentation technology in the prior art is prone to noise in the process of annotating colonoscopy images, resulting in low image segmentation accuracy.

[0076] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2 , Figure 2 A flow chart of the second embodiment of the colonoscopy image segmentation method of the present application.

[0077] In this embodiment, before step S10, the colonoscopy image segmentation method further includes steps S01 to S03:

[0078] Step S01: Input a plurality of colonoscopy training images into a preset image annotation model, so as to annotate each colonoscopy training image through the preset image annotation model, and obtain a plurality of annotated colonoscopy training images, wherein the annotated colonoscopy training images are annotated with corresponding image labels and confidence masks, and the confidence masks are used to identify noise areas in the image labels.

[0079] It should be understood that the above-mentioned colonoscopy training images may be colonoscopy images used to train the model, wherein this embodiment does not limit the number of colonoscopy training images.

[0080] It should be noted that the above image labels can be labels corresponding to each colonoscopy training image; correspondingly, the above confidence mask can be a value used to characterize the confidence level of a pixel, which can be used to identify the noise area in the scene training image. In this embodiment, if the confidence level of a pixel is high, its confidence mask can be marked as 1; if the confidence level of a pixel is low, the pixel is regarded as a pixel with edge noise, and its confidence mask can be marked as 0.

[0081] In practical applications, when training the preset image annotation model, we can first obtain unlabeled colon examination images from the server database and use the rich tools provided by the application to use different colors to clearly distinguish and accurately mark various types of ulcer areas, and after the annotation is completed, the annotation information is properly saved back to the server database. This annotation method enables this article to accurately mark the specific location and range of the ulcer, providing solid and valuable data support for subsequent model training. In order to better meet the requirements of model processing, this solution can convert the original annotation information into a fuzzy mask format. This fuzzy annotation is because when using the application for annotation, it is affected by factors such as device limitations, and there is inevitably noise in the annotation process, which can further improve the availability of data and the training effect of the model.

[0082] Furthermore, the step S01 includes:

[0083] Step S01a: inputting a plurality of colonoscopy training images into a preset image annotation model.

[0084] Step S01b: Determine the pixel distance between the target pixel and the corresponding target adjacent pixel in each colonoscopy training image by using the preset image annotation model.

[0085] It should be noted that the target pixel point may be any pixel point in the colonoscopy training image, and the target adjacent pixel point may be a pixel point around the target pixel point that is relatively close to the target pixel point. Accordingly, the pixel point distance is the distance between the target pixel point and the target adjacent pixel point.

[0086] Step S01c: determining the distance weight corresponding to the target pixel point based on the pixel point distance.

[0087] It can be understood that the distance weight mentioned above may be the proportion of the distance between pixels in evaluating the similarity between pixels.

[0088] Step S01d: Determine the similarity between the target pixel and the target adjacent pixel based on the distance weight.

[0089] In this embodiment, a difference function can be introduced to quantify the similarity between the jth target pixel and the kth adjacent pixel in the i-th colonoscopy training image: This similarity value can help determine the relationship between pixels and their common features. The difference function not only considers the value of the pixel when calculating the similarity, but also incorporates the weight coefficient And the exponential part of the difference between pixels. When calculating the confidence of a pixel, the distance weight It is determined based on the distance between pixels. Specifically, in, represents the distance between the jth pixel and the kth pixel in the i-th colonoscopy image (ie, the above-mentioned pixel distance). In this embodiment, such weight setting can consider the relative position relationship between the pixels, thereby helping to better capture the connection between adjacent pixels when calculating the similarity.

[0090] Step S01e: annotating each colonoscopy training image based on the similarity using the preset image annotation model to obtain a plurality of annotated colonoscopy training images.

[0091] Furthermore, the step S01e includes: determining the confidence of the target pixel based on the similarity; determining the confidence mask corresponding to the target pixel according to the confidence; and annotating each colonoscopy training image based on the similarity and the confidence mask by using the image annotation model to obtain a plurality of annotated colonoscopy training images.

[0092] It should be noted that the annotated colonoscopy training image may be an image obtained by annotating each pixel in the colonoscopy training image with its corresponding confidence mask.

[0093] In the specific implementation, after determining the similarity between the target pixel in the colonoscopy training image and its corresponding target adjacent pixel, we can further explore the label y i A specific pixel y ij This confidence depends not only on the similarity between the pixel and its surrounding pixels, but also on the possibility that the pixel is on the edge of the annotation.

[0094] Specifically, if the pixel y ij The higher the similarity with the surrounding pixels, the lower the possibility that the pixel is located at the edge of the annotation, and thus the higher its confidence. ij The confidence calculation formula is:

[0095] In the formula, Q ij Represents pixel y ij The confidence level, It represents the similarity between the jth pixel and the kth pixel in the i-th colonoscopy image, and μ is a parameter value used to determine the range of adjacent pixels considered.

[0096] This scheme compares the pixel y ij The similarity with the adjacent 2μ pixels can evaluate its confidence more comprehensively. This similarity-based weighted average method can comprehensively consider the local characteristics and spatial relationships of the pixels, thereby providing a more accurate confidence assessment.

[0097] After calculating the confidence of the target pixel, we can continue to calculate the confidence mask. Specifically, if m i Yes i The confidence mask of The confidence mask calculation method in this scheme is:

[0098]

[0099] In the formula, m i represents the confidence mask corresponding to the target pixel point, τ is a similarity threshold used to distinguish the confidence of the pixel point, and the value can be set according to the actual situation, and this embodiment does not limit this.

[0100] Specifically, if the pixel y ij The confidence level Q ij If the pixel y is greater than or equal to the threshold τ, the pixel is considered to have a higher confidence level and is marked as 1. ijIf the confidence is lower than the threshold τ, it is regarded as a pixel with edge noise and marked as 0. By setting an appropriate threshold, this scheme can better screen out pixels with higher confidence and exclude edge pixels that may contain noise. This process helps to improve the accuracy and stability of the segmentation results, thereby better coping with image data with complex backgrounds and noise.

[0101] In the specific implementation, refer to Figure 3 , Figure 3 This is a diagram of the architecture of the image segmentation model in the colonoscopy image segmentation method of this application. Figure 3 As shown in the figure, the model architecture proposed by this method consists of two core parts: a collaborative training segmentation method for colonoscopy images based on fuzzy annotation, and an intestinal ulcer annotation strategy based on active learning. In order to solve the problem that the existing model is prone to deviation in prediction due to direct use of noisy labeled images for training, this scheme can propose a collaborative training segmentation method based on annotation confidence. The model architecture proposed by this scheme consists of two core parts: annotation confidence region extraction and collaborative training segmentation model. Specifically, the annotation confidence region extraction module conducts in-depth analysis of the confidence information in the annotation data, so that the region with high confidence can be accurately identified and extracted; the collaborative training segmentation model is based on fuzzy annotation confidence and integrates the idea of ​​collaborative training framework, aiming to jointly solve the challenges caused by label ambiguity through mutual cooperation between multiple learners. This collaborative training method can not only enhance the generalization ability and robustness of the model, but also effectively utilize image data with noisy labels, thereby significantly improving the training efficiency of the segmentation model.

[0102] In this embodiment, if Figure 3 As shown, the annotation confidence region extraction module can first fuzzily annotate several colonoscopy training images in the public data set, and extract the confidence regions in these colonoscopy training images, that is, the regions with higher pixel confidence, and train the segmentation model based on the fuzzy annotation confidence by collaborative training, so that the training efficiency of the segmentation model can be significantly improved.

[0103] In this embodiment, refer to Figure 4 , Figure 4 This is a flow chart of the confidence region extraction module in the colonoscopy image segmentation method of this application. Figure 4 As shown in the figure, the model architecture proposed in this application consists of two core parts: the annotation confidence region extraction module and the collaborative training segmentation model. The mobile data annotation platform and IBD database can provide data support for the annotation confidence region extraction module. The annotation confidence region extraction module can calculate the confidence of the fuzzy label and calculate a confidence mask for each colonoscopy image. Among them, this mask can not only help this scheme identify the noise area in the label, but also provide uncertainty information of each pixel annotation for this method. The co-training segmentation model can be co-trained based on the semantic segmentation dataset consisting of images, fuzzy labels of images and confidence labels, which can not only enhance the generalization ability and robustness of the model, but also effectively utilize the image data with noise labels, thereby significantly improving the training efficiency of the segmentation model.

[0104] Step S02: constructing a semantic segmentation dataset based on the annotated colonoscopy training image, the image label and the confidence mask.

[0105] It should be noted that the above semantic segmentation dataset can be a dataset for training image segmentation models, which is composed of colonoscopy training images, labels corresponding to the images, and confidence masks. In this embodiment, after completing the calculation of the annotation confidence, a rich semantic segmentation dataset $\mathcal{D}$ can be obtained. Each image in the semantic segmentation dataset is accompanied by a corresponding label and confidence mask, which can serve as the basis for subsequent training and model optimization of this solution. In practical applications, the above semantic segmentation dataset contains N training samples, each of which consists of an image x i , label y i and confidence mask m i It consists of three parts, which can be expressed as These data will be used to train the model of this method to improve the learning ability of noisy labels.

[0106] Step S03: performing model training on the collaborative training segmentation model based on the semantic segmentation dataset to obtain a preset image segmentation model.

[0107] It should be noted that the above-mentioned collaborative training segmentation model can be a model that uses two parallel deep networks for data learning and classification.

[0108] Furthermore, the collaboratively trained segmentation model is composed of a first image segmentation network and a second image segmentation network; the first image segmentation network and the second image segmentation network both adopt a Unet++ architecture, in which a skip path and deep supervision are designed; the skip path is used to reduce the semantic gap between the feature maps of the encoder and decoder subnets, and the deep supervision is used to segment lesions.

[0109] In this embodiment, in order to improve the learning ability of noisy labels y, the present application can adopt a cooperative training framework (COT) strategy, which consists of two deep networks, namely the first image segmentation network and the second image segmentation network mentioned above, wherein the two networks are parallel, meaning that they run independently but share the same network architecture f. This design allows the present method to make full use of the information in the data set during training and enhance the robustness of the model to noisy labels.

[0110] It should be noted that the initial collaborative training segmentation model in this embodiment is constructed using all high-confidence pixels in the semantic segmentation dataset, which share the same architecture f. This design choice is to ensure that the initial structure of the network is highly relevant to the target segmentation task. High-confidence pixels are considered to have a lower noise level, so using them to build the network can provide a more accurate feature representation. At the same time, by sharing the same architecture f, the two networks can learn the basic knowledge of the segmentation task from the same foundation. In addition, the network architecture of the first image segmentation network and the second image segmentation network in this application is the Unet++ image segmentation network. The UNet++ architecture utilizes a redesigned skip path and deep supervision to achieve more accurate medical image segmentation. Among them, the redesigned skip path is intended to reduce the semantic gap between the feature maps of the encoder and decoder subnetworks, thereby solving a potentially simpler optimization problem for the optimizer, and deep supervision can also more accurately segment lesions, especially lesions that appear at multiple scales.

[0111] In this embodiment, the first image segmentation network and the second image segmentation network have independent parameters, denoted as θ1 and θ2 respectively. This parameter independence helps to avoid the model from falling into the local optimum in the early stage of training, thereby improving the generalization ability of the model. During initialization, the parameters of each network are initialized according to different random seeds to ensure that different networks have different starting points. The purpose of this is to encourage the two networks to explore different solution spaces during the training process, thereby improving the diversity and robustness of the model. In addition, when an image x is input i , each network will independently output the predicted log probability and

[0112] It should be noted that when dealing with image segmentation tasks with noisy labels, this solution faces a challenge, namely the waste of low-confidence pixel data. To solve this problem, the collaborative training framework aims to build two deep networks and run them in parallel to make full use of the information exchange and collaboration between the two networks. In this way, low-confidence labels can be better processed and the robustness and generalization ability of the model can be improved. Figure 5 , Figure 5 Schematic diagram of the collaborative training segmentation model in the colonoscopy image segmentation method of this application. Specifically, this application can label the pixels with low confidence (i.e., m ij =0) is considered as an uncertain label, m ij = 1 is considered a high confidence label. For high confidence pixels, use label y ij As supervision on both networks, it ensures that the networks can learn useful information from accurate labels. For pixels with low confidence, this method uses the prediction results of another network. (or ) as a guide. This design is because even if a pixel has a low confidence, the prediction result of another network at that pixel may have a high confidence. Introducing this collaborative mechanism can help the model better understand and handle noisy labels.

[0113] like Figure 5 As shown, in order to implement the above strategy, this scheme can define two loss functions:

[0114]

[0115] The two loss functions are used to optimize the first image segmentation network and the second image segmentation network, respectively. Specifically, the two loss functions calculate the cross entropy loss of each pixel and weight it according to the confidence mask of the pixel. Among them, if the confidence mask of the pixel is 1 (i.e. high confidence), the true label is used as the supervisory signal; if the confidence mask of the pixel is 0 (i.e. low confidence), the prediction result of the other network is used as a guide. In this way, it can be ensured that the two networks are consistent when dealing with uncertain labels, which helps to improve the robustness and generalization ability of the model, thereby achieving more accurate medical image segmentation in practical applications.

[0116] Finally, the total loss of joint training is:

[0117]

[0118] In this embodiment, it is disclosed that several colonoscopy training images are input into a preset image annotation model, so as to annotate each colonoscopy training image through the preset image annotation model to obtain several annotated colonoscopy training images, wherein the annotated colonoscopy training images are annotated with corresponding image labels and confidence masks, and the confidence mask is used to identify noise areas in the image labels; a semantic segmentation dataset is constructed based on the annotated colonoscopy training images, image labels and confidence masks; and a collaborative training segmentation model is trained based on the semantic segmentation dataset to obtain a preset image segmentation model. In this embodiment, the collaborative training method can not only enhance the generalization ability and robustness of the model, but also effectively utilize image data with noise labels, thereby significantly improving the training efficiency of the segmentation model.

[0119] Based on the first embodiment and / or the second embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the above embodiments can be referred to the above introduction, and will not be described in detail later. Figure 6 , Figure 6 A flow chart of the third embodiment of the colonoscopy image segmentation method of the present application.

[0120] In this embodiment, step S03 further includes steps S031 to S036:

[0121] Step S031: performing model training on the collaborative training segmentation model based on the semantic segmentation dataset to obtain an initial image segmentation model.

[0122] It should be understood that the above-mentioned initial image segmentation model can be an image segmentation model that has been preliminarily trained. In practical applications, after the initial model training is completed, the above-mentioned initial image segmentation model can be obtained. When it is applied to cross-center or private data sets, due to differences in data distribution, the so-called domain shift phenomenon, the performance of the model may be significantly affected. Therefore, in order to alleviate this problem and optimize the generalization ability of the model, the present application can introduce a comprehensive sampling strategy, in which the uncertainty and diversity of samples can be considered at the same time, and by combining the two, samples that are both challenging and informative can be screened out for labeling, so that the initial image segmentation model can continue to be iteratively trained through the labeled samples, thereby more effectively improving the performance of the model.

[0123] Step S032: Determine the uncertainty score corresponding to the sample colonoscopy image and the diversity score corresponding to the unlabeled sample colonoscopy image.

[0124] It should be understood that the above-mentioned sample colonoscopy images may be colonoscopy images in a hospital's private data set; correspondingly, the above-mentioned unlabeled sample colonoscopy images may be colonoscopy images that are not labeled in a hospital's private data set.

[0125] It should be noted that the above uncertainty score may be a value used to evaluate the uncertainty of the initial image segmentation model for the sample colonoscopy image; the above diversity score may be a value for measuring the diversity of the unlabeled sample colonoscopy images.

[0126] In practical applications, the uncertainty and diversity of samples can be considered simultaneously in the comprehensive sampling strategy to improve model performance and optimize the generalization ability of the model. In order to more effectively calculate the uncertainty score of the image, this method adopts the image entropy method. For a given sample image x i , this method can calculate its uncertainty score Score u (x i ) is used to evaluate the uncertainty of the model for the sample. The specific calculation formula is as follows:

[0127]

[0128] In the formula, Score u (x i ) represents the uncertainty score, W and H are the width and height of the feature map, respectively, and P ij represents the prediction result of the cth channel of the jth pixel in the i-th colonoscopy image, and C represents the number of classes. u (x i ) actually measures the degree of confusion or uncertainty in the model's prediction of the categories of each pixel in the image. If an image has a high uncertainty score, it means that the model's predictive ability for the image is relatively poor, so the image may contain new knowledge or information that the model has not yet learned.

[0129] In addition, in order to quantify the diversity of samples, this application can use a variety of distance measurement methods. Among them, cosine similarity is a commonly used method to measure the difference in direction between two vectors. For a given unlabeled sample This method uses the diversity score of the sample Defined as the distance between the sample and the labeled dataset. According to the k-Center-Greedy algorithm, the distance between an unlabeled point x and a set S is the minimum distance d2 between the point and each point in the set S. Therefore, the image diversity score of a sample can be the minimum distance to each labeled sample:

[0130] Step S033: determining a target sample colonoscopy image based on the uncertainty score and the diversity score.

[0131] It should be noted that the target sample colonoscopy images may be the first several sample images with the highest uncertainty and the most representative among the sample colonoscopy images and the unlabeled sample colonoscopy images.

[0132] Step S034: annotating the target sample colonoscopy image based on the expert annotation information to obtain an annotated sample colonoscopy image.

[0133] It should be understood that the above expert annotation information may be annotation information provided by experts in the field for images with high uncertainty.

[0134] Step S035: adding the labeled sample colonoscopy images to the original source data set to obtain an updated original source data set.

[0135] It should be noted that the above original source data set may be a public data set, such as an IBD database, etc., and this embodiment does not impose any limitation on this.

[0136] Step S036: iteratively training the initial image segmentation model based on the updated original source data set to obtain a preset image segmentation model.

[0137] In this embodiment, the image segmentation model can determine the most representative target sample colonoscopy images based on the evaluation results of sample uncertainty and diversity, and perform expert annotation on these target sample colonoscopy images based on expert annotation information to obtain annotated sample colonoscopy images, so as to maximize the model performance under limited annotation resources. The newly annotated samples are then added to In order to train a more accurate and robust model in the next round. Figure 3 As shown, the model architecture proposed in this application also includes a colonoscopy ulcer annotation model based on active data adaptation, wherein the colonoscopy ulcer annotation method based on active data adaptation is an iterative optimization process, which aims to gradually integrate the target dataset (i.e., the hospital private dataset) into the target dataset. ) to improve the performance of the model in the target domain. In this embodiment, the colonoscopy ulcer annotation method based on active data adaptation can be obtained from the hospital private dataset Filter out unlabeled data for expert annotation and add it to the original source dataset after the data is labeled The segmentation model is iteratively trained in .

[0138] In the specific implementation, refer to Figure 7 , Figure 7 This is a training flow chart of the preset image annotation model in the colonoscopy image segmentation method of this application. Figure 7 As shown, this scheme only uses the original source data set Starting from the initial model trained on Carefully select and label a certain number of samples, and then add these samples to In order to update and expand the source dataset. In each round of active learning, the current The trained model was used to evaluate its This step usually involves calculating uncertainty indicators such as prediction confidence or entropy of the sample. At the same time, in order to ensure that the selected samples are representative The algorithm also considers the diversity of samples, which can be evaluated by calculating the similarity between samples or using clustering methods. Based on these comprehensive evaluation results, the algorithm selects the top K samples with the highest uncertainty and the most representativeness for expert annotation, so as to maximize the model performance with limited annotation resources. After that, the newly annotated samples are subsequently added to In order to train a more accurate and robust image segmentation model in the next round.

[0139] In this embodiment, model training of a collaborative training segmentation model based on a semantic segmentation dataset is disclosed to obtain an initial image segmentation model; an uncertainty score corresponding to a sample colonoscopy image and a diversity score corresponding to an unlabeled sample colonoscopy image are determined; a target sample colonoscopy image is determined based on the uncertainty score and the diversity score; the target sample colonoscopy image is labeled based on expert annotation information to obtain a labeled sample colonoscopy image; the labeled sample colonoscopy image is added to the original source dataset to obtain an updated original source dataset; the initial image segmentation model is iteratively trained based on the updated original source dataset to obtain a preset image segmentation model, thereby screening out samples that are both challenging and informative for labeling, and the labeled samples are added to the original source dataset to iteratively train the image segmentation model, thereby more effectively improving the performance of the model.

[0140] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the colonoscopy image segmentation method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0141] This application also provides a colonoscopy image segmentation device, please refer to Figure 8 , the colonoscopy image segmentation device comprises:

[0142] A similarity determination module 10 is used to determine the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels;

[0143] A confidence determination module 20, configured to determine the pixel confidence corresponding to each pixel based on the pixel similarity;

[0144] An image annotation module 30, configured to perform confidence annotation on pixel labels in the colonoscopy image to be segmented based on the pixel confidence using a preset image annotation model to obtain an annotated colonoscopy image;

[0145] The image segmentation module 40 is used to segment the colonoscopic image to be segmented based on the annotated colonoscopic image by using a preset image segmentation model to obtain a colonoscopic segmented image.

[0146] The colonoscopic image segmentation device provided by the present application adopts the colonoscopic image segmentation method in the above embodiment, which can solve the technical problem that the intestinal ulcer segmentation technology in the prior art is prone to noise in the process of annotating colonoscopic images, resulting in low image segmentation accuracy. Compared with the prior art, the beneficial effects of the colonoscopic image segmentation device provided by the present application are the same as the beneficial effects of the colonoscopic image segmentation method provided by the above embodiment, and the other technical features of the colonoscopic image segmentation device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0147] The present application provides a colonoscopy image segmentation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the colonoscopy image segmentation method in the above-mentioned embodiment 1.

[0148] Reference below Fig. 9 , which shows a schematic diagram of the structure of a colonoscopy image segmentation device suitable for implementing the embodiment of the present application. The colonoscopy image segmentation device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig. 9 The colonoscopy image segmentation device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0149] like Fig. 9As shown, the colonoscopy image segmentation device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the colonoscopy image segmentation device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the colonoscopy image segmentation device to communicate with other devices wirelessly or by wire to exchange data. Although the colonoscopy image segmentation device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0150] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0151] The colonoscopic image segmentation device provided by the present application adopts the colonoscopic image segmentation method in the above embodiment, which can solve the technical problem of colonoscopic image segmentation. Compared with the prior art, the beneficial effects of the colonoscopic image segmentation device provided by the present application are the same as the beneficial effects of the colonoscopic image segmentation method provided by the above embodiment, and the other technical features in the colonoscopic image segmentation device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0152] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0153] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0154] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the colonoscopy image segmentation method in the above-mentioned embodiment.

[0155] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0156] The computer-readable storage medium may be included in the colonoscopy image segmentation device; or may exist independently without being assembled into the colonoscopy image segmentation device.

[0157] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the colonoscopic image segmentation device, the colonoscopic image segmentation device: determines the pixel similarity between all pixels in the colonoscopic image to be segmented and the corresponding adjacent pixels; determines the pixel confidence corresponding to each pixel based on the pixel similarity; performs confidence annotation on the pixel labels in the colonoscopic image to be segmented based on the pixel confidence through a preset image annotation model to obtain a labeled colonoscopic image; and performs segmentation on the colonoscopic image to be segmented based on the labeled colonoscopic image through a preset image segmentation model to obtain a colonoscopic segmented image.

[0158] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0159] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0160] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0161] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned colonoscopy image segmentation method, and can solve the technical problem that the intestinal ulcer segmentation technology in the prior art is prone to noise in the process of annotating colonoscopy images, resulting in low image segmentation accuracy. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the colonoscopy image segmentation method provided by the above-mentioned embodiment, and will not be repeated here.

[0162] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned colonoscopy image segmentation method when executed by a processor.

[0163] The computer program product provided by the present application can solve the technical problem that the intestinal ulcer segmentation technology in the prior art is prone to noise in the process of annotating colonoscopy images, resulting in low image segmentation accuracy. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the colonoscopy image segmentation method provided by the above embodiment, and will not be repeated here.

[0164] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A colonoscopy image segmentation method, characterized in that: The method includes: Determine the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels; Determine the pixel confidence corresponding to each pixel based on the pixel similarity; Performing confidence annotation on pixel labels in the colonoscopy image to be segmented based on the pixel confidence using a preset image annotation model to obtain an annotated colonoscopy image; The colonoscopy image to be segmented is segmented based on the annotated colonoscopy image using a preset image segmentation model to obtain a colonoscopy segmentation image.

2. The method according to claim 1, characterized in that Before the step of determining the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels, the method further includes: Inputting a plurality of colonoscopy training images into a preset image annotation model, so as to annotate each colonoscopy training image through the preset image annotation model, and obtaining a plurality of annotated colonoscopy training images, wherein the annotated colonoscopy training images are annotated with corresponding image labels and confidence masks, and the confidence masks are used to identify noise areas in the image labels; constructing a semantic segmentation dataset based on the annotated colonoscopy training image, the image label and the confidence mask; The collaboratively trained segmentation model is trained based on the semantic segmentation dataset to obtain a preset image segmentation model.

3. The method according to claim 2, characterized in that The step of inputting a plurality of colonoscopy training images into a preset image annotation model, annotating each colonoscopy training image by the preset image annotation model, and obtaining a plurality of annotated colonoscopy training images comprises: Input a number of colonoscopy training images into a preset image annotation model; Determine the pixel distance between the target pixel and the corresponding target adjacent pixel in each colonoscopy training image by using the preset image annotation model; Determine a distance weight corresponding to the target pixel point based on the pixel point distance; Determine the similarity between the target pixel and the target adjacent pixel based on the distance weight; Each colonoscopy training image is annotated based on the similarity using the preset image annotation model to obtain a plurality of annotated colonoscopy training images.

4. The method according to claim 3, characterized in that The step of annotating each colonoscopy training image based on the similarity by using the preset image annotation model to obtain a plurality of annotated colonoscopy training images includes: Determining the confidence of the target pixel based on the similarity; Determining a confidence mask corresponding to the target pixel according to the confidence level; The image annotation model is used to annotate each colonoscopy training image based on the similarity and the confidence mask to obtain a plurality of annotated colonoscopy training images.

5. The method according to claim 2, characterized in that The collaborative training segmentation model consists of a first image segmentation network and a second image segmentation network; both the first image segmentation network and the second image segmentation network adopt the Unet++ architecture, in which a skip path and deep supervision are designed; the skip path is used to reduce the semantic gap between the feature maps of the encoder and decoder subnets, and the deep supervision is used to segment lesions.

6. The method according to claim 2, characterized in that The step of performing model training on the collaborative training segmentation model based on the semantic segmentation data set to obtain a preset image segmentation model includes: Performing model training on the collaborative training segmentation model based on the semantic segmentation dataset to obtain an initial image segmentation model; Determine the uncertainty score corresponding to the sample colonoscopy image and the diversity score corresponding to the unlabeled sample colonoscopy image; Determine a target sample colonoscopy image based on the uncertainty score and the diversity score; Annotating the target sample colonoscopy image based on the expert annotation information to obtain an annotated sample colonoscopy image; Adding the labeled sample colonoscopy images to the original source data set to obtain an updated original source data set; The initial image segmentation model is iteratively trained based on the updated original source data set to obtain a preset image segmentation model.

7. A colonoscopy image segmentation device, characterized in that: The device comprises: A similarity determination module is used to determine the pixel similarity between all pixels in the colonoscopy image to be segmented and the corresponding adjacent pixels; A confidence determination module, used to determine the pixel confidence corresponding to each pixel based on the pixel similarity; An image annotation module, used to perform confidence annotation on pixel labels in the colonoscopy image to be segmented based on the pixel confidence using a preset image annotation model to obtain an annotated colonoscopy image; An image segmentation module is used to segment the colonoscopic image to be segmented based on the annotated colonoscopic image using a preset image segmentation model to obtain a colonoscopic segmented image.

8. A colonoscopy image segmentation device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the colonoscopy image segmentation method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the colonoscopy image segmentation method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the colonoscopy image segmentation method according to any one of claims 1 to 6 are implemented.

Citation Information

Patent Citations

  • Image segmentation network training and image segmentation method, device and equipment

    CN117095251A