Colorectal adenoma recognition semantic segmentation deep learning system and construction method

Abstract

The invention belongs to the technical field of learning model construction, and provides a colorectal adenoma recognition semantic segmentation deep learning system and a construction method, and the system comprises a data set obtaining and processing module which is used for obtaining and processing a case image of colorectal adenoma, and obtaining a target data set; the model construction module is used for constructing and obtaining Hamp on the basis of coloring agents named hematoxylin and eosin in combination with a semi-supervised learning model; e, a semantic segmentation deep learning model of adversarial average teachers; and the model learning training module is used for carrying out learning training on the semantic segmentation depth model based on the target data set. According to the method, the case image of the colorectal adenoma is acquired and processed, the semantic segmentation deep learning model is constructed, learned and trained on the basis of the coloring agent named hematoxylin and eosin and in combination with the semi-supervised learning model, the creation quality of the colorectal adenoma recognition semantic segmentation deep learning system can be ensured, and the recognition efficiency of the colorectal adenoma is improved. And a basis is provided for efficiently and accurately identifying the case image of the colorectal adenoma.

Description

Technical Field

[0001] The present invention relates to the technical field of learning model construction, and particularly to a colorectal adenoma recognition semantic segmentation deep learning system and a construction method thereof. Background Art

[0002] Histopathology provides the most accurate clinical treatment information for tumor diagnosis. However, even for experienced pathologists, manual evaluation can be time-consuming and subjective, which has led to the development of image analysis algorithms. Especially in recent years, deep learning-based methods have played an important role in the field of computational pathology.

[0003] Although deep convolutional neural networks have shown remarkable performance, the domain shift between training data (source domain) and test data (target domain) often hinders their generalization ability in real-world applications. In computational pathology, the variation in staining color is an important factor leading to staining; hematoxylin and eosin (H&E) is a popular stain applicable to whole-slide images, where the cell nuclei are blue (hematoxylin) and the cytoplasmic regions are pink (eosin). Multiple factors involved in the slide preparation, staining, and scanning processes can affect the final color appearance of colorectal adenoma case images, such as sample thickness, tissue fixation, staining protocol, scanning device, lighting conditions, etc. Therefore, slides from different medical institutions usually present different visual characteristics, and when deployed to a new hospital, the performance of deep convolutional neural networks often drops sharply;

[0004] To address this problem, existing methods aim to perform staining color normalization work, usually using image processing methods and deep generative models to segment the color distributions of training data and test data; and focusing on simulating the color variation of staining through color enhancement techniques; however, modifying the features of the input image does not always guarantee correct conversion and may introduce additional noise to the training stage.

[0005] Therefore, it is necessary to provide a colorectal adenoma recognition semantic segmentation deep learning system and a construction method thereof. Summary of the Invention

[0006] The present invention provides a colorectal adenoma recognition semantic segmentation deep learning system and a construction method thereof. By acquiring and processing colorectal adenoma case images, and based on a stain called hematoxylin and eosin (H&E), combined with a semi-supervised learning model, a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) is constructed, and the semantic segmentation deep model is learned and trained, which can ensure the creation quality of the colorectal adenoma recognition semantic segmentation deep learning system and provide a basis for efficiently and accurately identifying colorectal adenoma case images.

[0007] The present invention provides a semantic segmentation deep learning system for colorectal adenoma recognition, including:

[0008] A dataset acquisition and processing module, configured to acquire and process case images of colorectal adenomas to obtain a target dataset;

[0009] A model construction module, configured to construct a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) based on a stain called hematoxylin and eosin (H&E), in combination with a semi-supervised learning model (Adversarial Mean Teacher, AMT);

[0010] A model learning and training module, configured to perform learning and training on the semantic segmentation deep model based on the target dataset.

[0011] Further, the dataset acquisition and processing module includes a dataset acquisition unit and a dataset processing unit;

[0012] The dataset acquisition unit is configured to acquire case images of colorectal adenomas to form an original dataset;

[0013] The dataset processing unit is configured to process the original dataset to obtain a target dataset.

[0014] Further, processing the original dataset to obtain a target dataset includes:

[0015] Dividing the original dataset to obtain a training dataset and a test dataset;

[0016] Processing the training dataset to obtain a processed training dataset; wherein, the processing includes stain normalization processing and stain color enhancement using the HSV perturbation method; the stain normalization processing is performed to make its color distribution similar to that of the test dataset; the stain color enhancement is performed to achieve the expansion of the training dataset;

[0017] Summarizing the test dataset and the processed training dataset to form a target dataset.

[0018] Further, constructing a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) based on a stain called hematoxylin and eosin (H&E), in combination with a semi-supervised learning model (Adversarial Mean Teacher, AMT), includes:

[0019] Obtaining a teacher (Mean Teacher, MT) model and a student model;

[0020] Combining the student model and the teacher model to obtain a deep learning model; wherein, the teacher model is used to generate high-quality pseudo labels to supervise the learning of the student model in the target domain;

[0021] In the deep learning model, an aH&E discriminator with a Gradient Reversal Layer (GRL) is introduced for H&E-oriented adversarial learning to generate a semantic segmentation deep learning model that obtains the H&E Adversarial Mean Teacher (HEAMT). Among them, the adversarial learning is used to guide the semantic segmentation deep learning model to learn the stain-invariant representation through a min-max game.

[0022] Furthermore, the semantic segmentation deep model is trained, including:

[0023] Set the training dataset S and the test dataset T in the target dataset; where S includes N s case images of labeled colorectal adenomas, where, represents the i-th sample data in the training dataset S; T includes N t case images of unlabeled colorectal adenomas; where, represents the i-th sample data in the training dataset T;

[0024] According to the divided stages, based on the training dataset S and the test dataset T, optimize the loss in an end-to-end manner for learning and training; the stages include an initialization stage and a mutual learning stage;

[0025] Among them, the initialization stage is: training a fully supervised classification model based on the training dataset S, that is, training a classifier with the case images of labeled colorectal adenomas, and then copying it into two copies, and using the two copies as the initialized student model and teacher model respectively; the mutual learning stage is: the teacher model obtains weakly augmented images from the test dataset T, and the student model obtains strongly augmented images from the training dataset S and the test dataset T; the teacher model generates pseudo-labels on the test dataset T to train the student model, and the student model updates the teacher model through an exponential moving average.

[0026] Using the calculated H&E matrix, use an aH&E discriminator with a Gradient Reverse Layer (GRL) to align the stain distribution in the feature space.

[0027] Furthermore, the mutual learning stage also includes:

[0028] Using strong augmentation strategies and weak augmentation strategies to help the student model learn stain-invariant features in the training dataset S and the test dataset T;

[0029] Introduce reverse learning into the semantic segmentation deep learning model to align the distributions of the training dataset S and the test dataset T in the feature space.

[0030] Furthermore, the weak augmentation strategy includes random horizontal flipping, random vertical flipping, and random rotation; the strong augmentation strategy adds HSV shifting on the basis of the weak augmentation strategy.

[0031] Furthermore, using the calculated H&E matrix, an aH&E discriminator with a Gradient Reversal Layer (GRL) is used to align the staining distribution in the feature space, including:

[0032] Improve the obtained discriminator to obtain an improved discriminator for predicting H&E components;

[0033] Change the binary cross-entropy adversarial loss to:

[0034]

[0035] where L adv represents the adversarial loss, and L MSE represents the square of the loss function L2; represents the H&E matrix of the image in the strong augmentation training dataset S; represents the H&E matrix of the image in the strong augmentation test dataset T;

[0036] Based on the improved discriminator, apply the binary cross-entropy adversarial loss and combine the Gradient Reversal Layer (GRL) between classification networks to achieve adversarial learning.

[0037] Furthermore, the student model updates the teacher model through the exponential moving average, including:

[0038] The student model is updated through the backpropagation algorithm of gradient descent, and the weights of the teacher model are refined through the exponential moving average (EMA), specifically:

[0039]

[0040] In the above formula, represents the teacher model, t represents the number of iterations, and α represents the smoothing constant.

[0041] A construction method for constructing a deep learning system for semantic segmentation of colorectal adenoma recognition, including:

[0042] Obtain and process the case images of colorectal adenomas to obtain the target dataset;

[0043] Based on the stain called hematoxylin and eosin (H&E), combined with the semi-supervised learning model (Adversarial MeanTeacher, AMT), construct a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT);

[0044] Based on the target dataset, the semantic segmentation deep model is learned and trained.

[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects: By acquiring and processing the case images of colorectal adenomas, and based on a stain called hematoxylin and eosin (H&E), combined with a semi-supervised learning model, a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) is constructed, and the semantic segmentation deep model is learned and trained, which can ensure the creation quality of the semantic segmentation deep learning system for colorectal adenoma recognition, and provide a basis for efficiently and accurately identifying the case images of colorectal adenomas.

[0046] Other features and advantages of the present invention will be described in the following specification, and part of them will be obvious from the specification, or understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification and the accompanying drawings.

[0047] The technical solutions of the present invention will be further described in detail below through the accompanying drawings and embodiments. Description of the Drawings

[0048] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0049] Figure 1 It is a schematic structural diagram of a semantic segmentation deep learning system for colorectal adenoma recognition;

[0050] Figure 2 It is a schematic structural diagram of the dataset acquisition and processing module;

[0051] Figure 3 It is a schematic diagram of the method steps of a construction method. Detailed Embodiments

[0052] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0053] The present invention provides a semantic segmentation deep learning system for colorectal adenoma recognition, as Figure 1 shown, including:

[0054] A dataset acquisition and processing module, configured to acquire and process the case images of colorectal adenomas to obtain a target dataset;

[0055] A model construction module, which is used to construct a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) based on a stain called hematoxylin and eosin (H&E) and in combination with a semi-supervised learning model (Adversarial Mean Teacher, AMT);

[0056] A model learning and training module, which is used to learn and train the semantic segmentation deep model based on the target dataset.

[0057] The working principle of the above technical solution is: In order to build a semantic segmentation deep learning system for colorectal adenoma recognition, the present invention proposes a dataset acquisition and processing module, which is used to acquire and process the case images of colorectal adenoma to obtain a target dataset; a model construction module, which is used to construct a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) based on a stain called hematoxylin and eosin (H&E) and in combination with a semi-supervised learning model (Adversarial Mean Teacher, AMT); a model learning and training module, which is used to learn and train the semantic segmentation deep model based on the target dataset.

[0058] The beneficial effect of the above technical solution is: By adopting the solution provided in this embodiment, by acquiring and processing the case images of colorectal adenoma, and based on a stain called hematoxylin and eosin (H&E), in combination with a semi-supervised learning model, constructing a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT), and learning and training the semantic segmentation deep model, the creation quality of the semantic segmentation deep learning system for colorectal adenoma recognition can be guaranteed, providing a basis for efficiently and accurately recognizing the case images of colorectal adenoma.

[0059] In one embodiment, as Figure 2 shown, the dataset acquisition and processing module includes a dataset acquisition unit and a dataset processing unit;

[0060] The dataset acquisition unit is used to acquire the case images of colorectal adenoma to form an original dataset;

[0061] The dataset processing unit is used to process the original dataset to obtain a target dataset.

[0062] The working principle of the above technical solution is: In order to realize the acquisition and processing of the dataset, the present invention first acquires the case images of colorectal adenoma to form an original dataset; then processes the original dataset to obtain a target dataset.

[0063] The beneficial effect of the above technical solution is: By adopting the solution provided in this embodiment, through the acquisition and processing of the original dataset, a data basis is provided for the construction of the subsequent learning model.

[0064] In one embodiment, the original dataset is processed to obtain a target dataset, including:

[0065] The original dataset is divided to obtain a training dataset and a test dataset;

[0066] The training dataset is processed to obtain a processed training dataset; wherein, the processing includes stain normalization processing and stain color enhancement using the HSV perturbation method; the stain normalization processing is performed to make its color distribution similar to that of the test dataset; the stain color enhancement is performed to achieve the expansion of the training dataset;

[0067] The test dataset and the processed training dataset are aggregated to form the target dataset.

[0068] The working principle of the above technical solution is as follows: In order to process the original dataset, the present invention first divides the original dataset to obtain a training dataset and a test dataset; then processes the training dataset to obtain a processed training dataset; wherein, the processing includes stain normalization processing and stain color enhancement using the HSV perturbation method; the stain normalization processing is performed to make its color distribution similar to that of the test dataset; the stain color enhancement is performed to achieve the expansion of the training dataset; finally, the test dataset and the processed training dataset are aggregated to form the target dataset.

[0069] The beneficial effect of the above technical solution is: By adopting the solution provided in this embodiment, through the processing of the original dataset, the usability of data processing can be ensured, so as to enhance the learning and training effect of the learning model in learning and training.

[0070] In one embodiment, based on a stain called hematoxylin and eosin (H&E), combined with a semi-supervised learning model (Adversarial Mean Teacher, AMT), a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) is constructed, including:

[0071] Obtain a teacher (Mean Teacher, MT) model and a student model;

[0072] The student model and the teacher model are combined to obtain a deep learning model; wherein, the teacher model is used to generate high-quality pseudo-labels to supervise the learning of the student model in the target domain;

[0073] In the deep learning model, an aH&E discriminator with a gradient reversal layer (GRL) is introduced to perform H&E-oriented adversarial learning, and a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT) is generated, wherein the adversarial learning is used to guide the semantic segmentation deep learning model to learn stain-invariant representations through a min-max game.

[0074] The working principle of the above technical solution is as follows: In the process of constructing a semantic segmentation deep learning model for obtaining the H&E adversarial mean teacher (HEAMT), the present invention first obtains a teacher (Mean Teacher, MT) model and a student model; then combines the student model and the mean teacher model to obtain a deep learning model; among them, the teacher model is used to generate high-quality pseudo-labels to supervise the learning of the student model in the target domain; in the deep learning model, an aH&E discriminator with a gradient reversal layer (GRL) is introduced to perform H&E-oriented adversarial learning to generate a semantic segmentation deep learning model for obtaining the H&E adversarial mean teacher (HEAMT), where the adversarial learning is used to guide the semantic segmentation deep learning model to learn color invariant representations through a min-max game.

[0075] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, by obtaining a teacher model and a student model, combining the student model and the mean teacher model to obtain a deep learning model, and finally introducing an aH&E discriminator with a gradient reversal layer (GRL) to perform H&E-oriented adversarial learning, a high-quality semantic segmentation deep learning model can be guaranteed.

[0076] In one embodiment, the learning and training of the semantic segmentation deep model includes:

[0077] Set the training dataset S and the test dataset T in the target dataset; where S includes N s case images of labeled colorectal adenomas, where represents the i-th sample data in the training dataset S; T includes N t case images of unlabeled colorectal adenomas; where represents the i-th sample data in the training dataset T;

[0078] According to the divided stages, based on the training dataset S and the test dataset T, optimize the loss in an end-to-end manner for learning and training; the stages include an initialization stage and a mutual learning stage;

[0079] Among them, the initialization stage is: training a fully supervised classification model based on the training dataset S, that is, training a classifier with the case images of labeled colorectal adenomas, and then copying it into two copies, and using the two copies as the initialized student model and teacher model respectively; the mutual learning stage is: the teacher model obtains weakly augmented images from the test dataset T, and the student model obtains strongly augmented images from the training dataset S and the test dataset T; the teacher model generates pseudo-labels on the test dataset T to train the student model, and the student model updates the teacher model through an exponential moving average.

[0080] Using the calculated H&E matrix, an aH&E discriminator with a gradient reversal layer (GRL) is used to align the staining distribution in the feature space.

[0081] The working principle of the above technical solution is as follows: To implement the learning and training of the semantic segmentation deep model, the present invention first sets a training data set S and a test data set T in the target data set; where S includes N s case images of labeled colorectal adenomas, where represents the i-th sample data in the training data set S; T includes N t case images of unlabeled colorectal adenomas; where represents the i-th sample data in the training data set T; then, according to the divided stages, based on the training data set S and the test data set T, the loss is optimized in an end-to-end manner for learning and training; the stages include an initialization stage and a mutual learning stage; where the initialization stage is: training a fully supervised classification model based on the training data set S, that is, training a classifier with the case images of labeled colorectal adenomas, and then copying it into two copies, and using the two copies as the initialized student model and teacher model respectively; the mutual learning stage is: the teacher model obtains weakly augmented images from the test data set T, and the student model obtains strongly augmented images from the training data set S and the test data set T; the teacher model generates pseudo-labels on the test data set T to train the student model, and the student model updates the teacher model through an exponential moving average; finally, using the calculated H&E matrix, an aH&E discriminator with a gradient reversal layer (GRL) is used to align the staining distribution in the feature space.

[0082] The beneficial effect of the above technical solution is: By adopting the solution provided in this embodiment, the quality of learning and training can be guaranteed by implementing the learning and training of the semantic segmentation deep model.

[0083] In one embodiment, the mutual learning stage further includes:

[0084] Using strong augmentation strategies and weak augmentation strategies to help the student model learn stain-invariant features in the training data set S and the test data set T;

[0085] Introducing inverse learning into the semantic segmentation deep learning model to align the distributions of the training data set S and the test data set T in the feature space.

[0086] The working principle of the above technical solution is as follows: The mutual learning stage in the present invention further includes: using strong augmentation strategies and weak augmentation strategies to help the student model learn color-invariant features in the training dataset S and the test dataset T; introducing inverse learning into the deep learning model for semantic segmentation to align the distributions of the training dataset S and the test dataset T in the feature space.

[0087] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, through learning color-invariant features and introducing inverse learning into the deep learning model for semantic segmentation, the regularity of the feature distributions of the training dataset S and the test dataset T can be ensured.

[0088] In one embodiment, the weak augmentation strategies include random horizontal flipping, random vertical flipping, and random rotation; the strong augmentation strategies add HSV shifting on the basis of the weak augmentation strategies.

[0089] The working principle of the above technical solution is as follows: The weak augmentation strategies in the present invention include random horizontal flipping, random vertical flipping, and random rotation; the strong augmentation strategies add HSV shifting on the basis of the weak augmentation strategies.

[0090] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, through the combined use of weak augmentation strategies and strong augmentation strategies, color-invariant features can be ensured to be learned.

[0091] In one embodiment, using the calculated H&E matrix, aligning the color distributions in the feature space using an aH&E discriminator with a gradient reversal layer (GRL) includes:

[0092] Improving the obtained discriminator to obtain an improved discriminator for predicting H&E components;

[0093] Changing the binary cross-entropy adversarial loss to:

[0094]

[0095] where L adv represents the adversarial loss, and L MSE represents the square of the loss function L2; represents the H&E matrix of the image in the strongly augmented training dataset S; represents the H&E matrix of the image in the strongly augmented test dataset T;

[0096] Based on the improved discriminator, applying the binary cross-entropy adversarial loss and combining the gradient reversal layer (GRL) between classification networks to achieve adversarial learning.

[0097] The working principle of the above technical solution is as follows: Due to the lack of annotations for the target data, pseudo-labels in the teacher model are used for self-training. However, since the supervision in the initialization stage only comes from the training dataset S, the initial model at the beginning of the mutual learning stage may have a strong source bias, and this domain bias needs to be compensated; Traditional adversarial learning involves two important parts: a feature extractor and a domain discriminator; The domain discriminator is used to distinguish features from the training dataset S or the test dataset T and predict the probability of the test dataset T for each sample, and then the binary cross-entropy adversarial loss Ladv is applied to the label image based on the probability; In order to utilize the calculated H&E matrix, an aH&E discriminator with a gradient reversal layer (GRL) is used to align the staining distribution in the feature space. In the present invention, the obtained discriminator is first improved to obtain an improved discriminator for predicting the H&E components; Then the binary cross-entropy adversarial loss is changed, and then, based on the improved discriminator, the binary cross-entropy adversarial loss is applied, and the gradient reversal layer (GRL) between the classification networks is combined to achieve adversarial learning.

[0098] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, through improving the obtained discriminator and then changing the binary cross-entropy adversarial loss, visual representations irrelevant to staining changes can be learned, so as to align images of different colors into a common space. Then, based on the improved discriminator, the binary cross-entropy adversarial loss is applied, and the gradient reversal layer (GRL) between the classification networks is combined to achieve adversarial learning, which can ensure the quality of adversarial learning.

[0099] In one embodiment, the student model updates the teacher model through an exponential moving average, including:

[0100] The student model is updated through the backpropagation algorithm of gradient descent, and the weights of the teacher model are refined through the exponential moving average (EMA), specifically:

[0101]

[0102] In the above formula, represents the teacher model, t represents the number of iterations, and α represents the smoothing constant.

[0103] The working principle of the above technical solution is as follows: In the present invention, the student model updates the teacher model through the exponential moving average, which is reflected in that the student model is updated through the backpropagation algorithm of gradient descent, and the weights of the teacher model are refined through the exponential moving average (EMA).

[0104] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, the teacher model can be regarded as a collection of learning models of different versions during the entire pre-training process. Therefore, the robustness and stability of the teacher model are guaranteed, and accurate pseudo-labels can be generated.

[0105] A construction method, as Figure 3 shown, for constructing a semantic segmentation deep learning system for colorectal adenoma recognition, including:

[0106] Obtain and process the case images of colorectal adenomas to obtain a target data set;

[0107] Based on a staining agent called hematoxylin and eosin (H&E), combined with a semi-supervised learning model (Adversarial MeanTeacher, AMT), construct a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT);

[0108] Based on the target data set, perform learning and training on the semantic segmentation deep model.

[0109] The working principle of the above technical solution is as follows: To implement the construction method, the present invention first obtains and processes the case images of colorectal adenomas to obtain a target data set; then, based on a staining agent called hematoxylin and eosin (H&E), combined with a semi-supervised learning model (Adversarial Mean Teacher, AMT), constructs a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT); finally, based on the target data set, performs learning and training on the semantic segmentation deep model.

[0110] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, by obtaining and processing the case images of colorectal adenomas, and based on a staining agent called hematoxylin and eosin (H&E), combined with a semi-supervised learning model, constructing a semantic segmentation deep learning model of H&E adversarial mean teacher (HEAMT), and performing learning and training on the semantic segmentation deep model, the creation quality of the semantic segmentation deep learning system for colorectal adenoma recognition can be guaranteed, providing a basis for efficiently and accurately recognizing the case images of colorectal adenomas.

[0111] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A colorectal adenoma recognition semantic segmentation deep learning system, characterized in that: include: A data set acquisition and processing module is used to acquire and process case images of colorectal adenoma to obtain a target data set; A model building module for building a semantic segmentation deep learning model for obtaining the H&E Adversarial Mean Teacher (HEAMT) based on a stain called hematoxylin and eosin (H&E) combined with a semi-supervised learning model (Adversarial Mean Teacher, AMT); The model learning and training module is used to learn and train the semantic segmentation deep model based on the target dataset.

2. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 1, characterized in that: The data set acquisition and processing module includes a data set acquisition unit and a data set processing unit; A data set acquisition unit, used for acquiring case images of colorectal adenoma to form an original data set; The data set processing unit is used to process the original data set to obtain the target data set.

3. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 2, characterized in that: Process the original data set to obtain the target data set, including: Divide the original data set to obtain a training data set and a test data set; The training data set is processed to obtain a processed training data set; wherein the processing includes color normalization processing and color enhancement using the HSV perturbation method; the color normalization processing is performed to make it have a color distribution similar to that of the test data set; the color enhancement is performed to achieve the expansion of the training data set; Aggregate the test data set and the processed training data set to form the target data set.

4. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 1, characterized in that: Based on a dye called hematoxylin and eosin (H&E), combined with a semi-supervised learning model (Adversarial Mean Teacher, AMT), a semantic segmentation deep learning model for obtaining H&E Adversarial Mean Teacher (HEAMT) is constructed, including: Get the teacher (Mean Teacher, MT) model and the student model; Combine the student model and the teacher model to obtain a deep learning model. The teacher model is used to generate high-quality pseudo labels to supervise the student model's learning in the target domain. In the deep learning model, aH&E discriminator with a gradient reversal layer (GRL) is introduced to perform H&E oriented adversarial learning, generating a semantic segmentation deep learning model with the H&E Adversarial Average Teacher (HEAMT), where adversarial learning is used to guide the semantic segmentation deep learning model to learn coloring invariant representation through a min-max game.

5. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 1, characterized in that: Based on the target dataset, the semantic segmentation deep model is trained, including: Set the training dataset S and test dataset T in the target dataset; where S includes N s Case images of a marked colorectal adenoma, in, Represents the i-th sample data in the training data set S; T includes N t Case images of unlabeled colorectal adenomas; in, Represents the i-th sample data in the training data set T; According to the divided stages, based on the training data set S and the test data set T, the loss is optimized in an end-to-end manner to perform learning and training; the stages include the initialization stage and the mutual learning stage; Among them, the initialization stage is: training a fully supervised classification model based on the training data set S, that is, training the classifier with labeled case images of colorectal adenoma, and then copying it into two copies, and using the two copies as the initialized student model and teacher model respectively; the mutual learning stage is: the teacher model obtains weakly enhanced images from the test data set T, and the student model obtains strongly enhanced images from the training data set S and the test data set T; the teacher model generates pseudo labels on the test data set T to train the student model, and the student model updates the teacher model through the exponential moving average; Using the calculated H&E matrix, aH&E discriminator with gradient reversal layer (GRL) was used to align the staining distribution in feature space.

6. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 5, characterized in that: The mutual learning phase also includes: Using strong and weak enhancement strategies, the student model is helped to learn coloring invariant features in the training dataset S and the test dataset T; Inverse learning is introduced into the semantic segmentation deep learning model to align the distribution of the training dataset S and the test dataset T in the feature space.

7. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 6, characterized in that: The weak enhancement strategy includes random horizontal flipping, random vertical flipping and random rotation; the strong enhancement strategy adds HSV shift based on the weak enhancement strategy.

8. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 6, characterized in that: Using the calculated H&E matrix, aH&E discriminator with gradient reversal layer (GRL) is used to align the staining distribution in feature space, including: improving the obtained discriminator to obtain an improved discriminator for predicting H&E components; Change the binary cross entropy adversarial loss to: Among them, L adv represents the adversarial loss, L MSE Represents the square of the loss function L2; Represents the image in the strongly enhanced training dataset S H&E matrix; Represents the image in the strongly enhanced test dataset T H&E matrix; Based on the improved discriminator, binary cross entropy adversarial loss is applied, and the gradient reversal layer (GRL) between the classification networks is combined to achieve adversarial learning.

9. A colorectal adenoma recognition semantic segmentation deep learning system according to claim 6, characterized in that: The student model updates the teacher model through an exponential moving average, including: The student model is updated by the back-propagation algorithm of gradient descent, and the weights of the teacher model are refined by the exponential moving average (EMA), specifically: In the above formula, represents the teacher model, t represents the number of iterations, and α represents the smoothing constant.

10. A construction method, characterized in that: Used to construct a colorectal adenoma recognition semantic segmentation deep learning system as described in any one of claims 1 to 9, characterized by comprising: Acquire and process case images of colorectal adenoma to obtain the target data set; Based on a stain called hematoxylin and eosin (H&E), combined with a semi-supervised learning model (Adversarial MeanTeacher, AMT), a semantic segmentation deep learning model for obtaining H&E Adversarial Mean Teacher (HEAMT) was constructed; Based on the target dataset, the semantic segmentation deep model is learned and trained.

Images