A deep learning-based pathological image batch effect optimization method and system

Abstract

The application discloses a kind of pathological image batch effect optimization method and system based on deep learning, which comprises the following steps: obtaining pathological sample images and performing image preprocessing to obtain multi-center pathological image data to be processed;Batch discriminator training module and batch contrast module are introduced to construct batch effect optimization neural network model;Based on batch effect optimization neural network model, the multi-center pathological image data to be processed is subjected to pathological image batch effect optimization processing to obtain optimized multi-center pathological image data.By using the application, the batch effect of multi-center pathological image data can be reduced, and the adaptability and prediction accuracy of the model to multiple center data can be improved.The application is a kind of pathological image batch effect optimization method and system based on deep learning, which can be widely applied in the field of image recognition technology.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for optimizing batch effects in pathological images based on deep learning. Background Technology

[0002] Current pathological diagnosis primarily relies on pathologists manually examining H&E-stained tissue sections under microscopes. While widely used clinically, this method is inefficient and its results are highly dependent on the operator's skill and experience, making it prone to errors. With the development of digital pathology, the use of full-frame digital pathology images (WSI) has become widespread, enabling the application of artificial intelligence technologies such as deep learning. These technologies can automate image analysis, improving diagnostic efficiency and consistency. However, data from multicenter studies often suffer from batch effects due to differences in acquisition protocols, equipment manufacturers, and equipment drift between centers. These effects cause performance degradation of models built during training when applied to data from new centers, limiting the model's generalization ability. While current data preprocessing methods such as gradient distortion correction, bias field correction, and intensity normalization alleviate batch effects to some extent, they do not fundamentally solve the problem. Furthermore, existing processing methods typically rely on statistical tests or machine learning algorithms, which are highly dependent on the distribution and characteristics of the data and are not effective in all situations. Moreover, existing models are usually trained and optimized on specific datasets, but in real-world applications, especially in cross-center application scenarios, there are significant changes in data distribution and conditions, and existing models often struggle to adapt to these changes. Summary of the Invention

[0003] To address the aforementioned technical problems, the present invention aims to provide a method and system for optimizing batch effects in pathological images based on deep learning. This method can reduce the batch effect of multi-center pathological image data, thereby improving the model's adaptability and prediction accuracy to multi-center data.

[0004] The first technical solution adopted in this invention is: a method for optimizing batch effects in pathological images based on deep learning, comprising the following steps:

[0005] Acquire pathological sample images and perform image preprocessing to obtain multicenter pathological image data to be processed;

[0006] A batch discriminator training module and a batch adversarial module are introduced to construct a batch effect optimization neural network model;

[0007] The batch effect optimization neural network model is used to perform batch effect optimization processing on the multicenter pathological image data to be processed, resulting in optimized multicenter pathological image data.

[0008] Furthermore, the step of acquiring pathological sample images and performing image preprocessing to obtain multicenter pathological image data to be processed specifically includes:

[0009] Image data extraction and processing are performed based on pathological sample data to obtain pathological sample images;

[0010] The pathological sample images are converted into digital format using a digital pathological image scanner.

[0011] The pathological sample images in digital format are sequentially processed by image enhancement, threshold segmentation, and target region cutting to obtain the multicenter pathological image data to be processed.

[0012] Furthermore, the batch effect optimized neural network model includes a data processing module, a batch discriminator training module, and a batch adversarial module. The first output of the data processing module is connected to the first input of the batch discriminator training module, the second output of the data processing module is connected to the first input of the batch adversarial module, the output of the batch discriminator training module is connected to the second input of the batch adversarial module, and the output of the batch adversarial module is connected to the second input of the batch discriminator training module.

[0013] Furthermore, the step of performing batch effect optimization on the multicenter pathological image data to be processed using the batch effect optimization neural network model to obtain optimized multicenter pathological image data specifically includes:

[0014] Determine batch training labels based on the multicenter pathological image data to be processed;

[0015] The multicenter pathological image data to be processed and the batch training labels are input into the batch effect optimization neural network model;

[0016] The data processing module based on the batch effect optimized neural network model performs slicing and feature extraction on the multicenter pathological image data to be processed, and obtains multicenter pathological image feature vector data.

[0017] The batch discriminator training module based on the batch effect optimization neural network model performs batch prediction on the feature vector data of multicenter pathological images and batch training labels to obtain the batch prediction probability value of multicenter pathological images.

[0018] The batch adversarial module based on the batch effect optimizes the neural network model and performs backpropagation to update the batch prediction probability values ​​and batch training labels of multicenter pathological images, thereby obtaining optimized multicenter pathological image data.

[0019] Furthermore, the batch discriminator training module based on the batch effect optimized neural network model, in the step of performing batch prediction on the multi-center pathological image feature vector data and batch training labels to obtain the batch prediction probability value of the multi-center pathological image, specifically includes:

[0020] Multicenter pathological image feature vector data and batch training labels are input into a batch discriminator training module based on a batch effect optimized neural network model. The batch discriminator training module includes a first linear transformation layer, a first nonlinear transformation layer, a second linear transformation layer, and a second nonlinear transformation layer.

[0021] Based on the first linear transformation layer of the batch discriminator training module, the feature dimension of the multicenter pathological image feature vector data is halved to obtain the linearly transformed multicenter pathological image feature vector data.

[0022] Based on the first nonlinear transformation layer of the batch discriminator training module, a nonlinear transformation is performed on the feature vector data of the multicenter pathological image after linear transformation to obtain the feature vector data of the multicenter pathological image after nonlinear transformation.

[0023] Based on the second linear transformation layer of the batch discriminator training module, linear transformation processing is performed on the feature vector data of multicenter pathological images after nonlinear transformation to obtain the batch prediction probability of multicenter pathological images.

[0024] Based on the second nonlinear transformation layer of the batch discriminator training module, nonlinear transformation and confidence conversion are performed on the batch prediction probability and batch training label of multicenter pathological images to obtain the batch prediction probability value of multicenter pathological images.

[0025] Furthermore, the batch adversarial module based on the batch effect optimization neural network model, which updates the batch prediction probability values ​​and batch training labels of multi-center pathological images through backpropagation to obtain optimized multi-center pathological image data, specifically includes:

[0026] Invert the batch training labels to obtain the inverted batch training labels.

[0027] The inverted batch training labels and the batch prediction probability values ​​of multicenter pathological images are input into the batch adversarial module based on the batch effect optimization neural network model to calculate the batch adversarial loss and obtain the batch loss value of multicenter pathological images.

[0028] The batch discriminator training module is updated by backpropagation based on the batch loss value of multicenter pathological images to obtain the updated batch discriminator training module.

[0029] Based on the updated batch discriminator training module, optimized multicenter pathological image data is output.

[0030] Furthermore, the batch discriminator training module employs a topkpatch selection mechanism and an early stop training strategy to improve the data processing efficiency of the batch discriminator training module.

[0031] The second technical solution adopted in this invention is: a deep learning-based batch effect optimization system for pathological images, comprising:

[0032] The first module is used to acquire pathological sample images and perform image preprocessing to obtain multicenter pathological image data to be processed.

[0033] The second module is used to introduce the batch discriminator training module and the batch adversarial module to build a batch effect optimization neural network model.

[0034] The third module is used to perform batch effect optimization processing on the multicenter pathological image data to be processed based on the batch effect optimization neural network model, so as to obtain the optimized multicenter pathological image data.

[0035] The beneficial effects of the method and system of this invention are as follows: This invention obtains multi-center pathological image data to be processed by acquiring pathological sample images and performing image preprocessing. Then, it introduces a batch discriminator training module and a batch adversarial module to construct a batch effect optimization neural network model. Finally, based on the batch effect optimization neural network model, it performs pathological image batch effect optimization processing on the multi-center pathological image data to be processed. By introducing the batch discriminator training module and the batch adversarial module, the discrimination effect of the batch discriminator is enhanced by using extracted features and batch labels. The batch adversarial module is applied to optimize and train the entire network to reduce the batch effect and improve the model's adaptability and prediction accuracy for multi-center data. Attached Figure Description

[0036] Figure 1 This is a flowchart of the steps of a deep learning-based method for optimizing batch effects in pathological images according to the present invention.

[0037] Figure 2 This is a structural block diagram of a deep learning-based batch effect optimization system for pathological images according to the present invention.

[0038] Figure 3 This is a schematic diagram of a batch effect optimization neural network model provided in a specific embodiment of the present invention. Detailed Implementation

[0039] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0040] Reference Figure 1 This invention provides a deep learning-based method for optimizing batch effects in pathological images, which includes the following steps:

[0041] S100: Acquire pathological sample images and perform image preprocessing to obtain multicenter pathological image data to be processed;

[0042] Specifically, image data extraction and processing are performed based on pathological sample data to obtain pathological sample images; the pathological sample images are then converted to digital format using a digital pathological image scanner to obtain pathological sample images with digital format; image enhancement, threshold segmentation, and target region cutting are then performed sequentially on the pathological sample images with digital format to obtain multi-center pathological image data to be processed.

[0043] In this embodiment, after images are acquired from pathological samples, they are converted into digital format by a digital pathological image scanner and transmitted to a computing platform. The software running on the computing platform then performs image preprocessing, including image enhancement, threshold segmentation, and target region cutting.

[0044] S200: Introduce a batch discriminator training module and a batch adversarial module to construct a batch effect optimized neural network model;

[0045] Specifically, such as Figure 3 As shown, the batch effect optimized neural network model includes a data processing module, a batch discriminator training module, and a batch adversarial module. The first output of the data processing module is connected to the first input of the batch discriminator training module, the second output of the data processing module is connected to the first input of the batch adversarial module, the output of the batch discriminator training module is connected to the second input of the batch adversarial module, and the output of the batch adversarial module is connected to the second input of the batch discriminator training module.

[0046] S300. Based on the batch effect optimization neural network model, the multicenter pathological image data to be processed is subjected to batch effect optimization processing to obtain optimized multicenter pathological image data.

[0047] Specifically, batch training labels are determined based on the multicenter pathological image data to be processed; the multicenter pathological image data to be processed and the batch training labels are input into a batch effect optimization neural network model; the data processing module based on the batch effect optimization neural network model performs slicing and feature extraction processing on the multicenter pathological image data to be processed to obtain multicenter pathological image feature vector data; the batch discriminator training module based on the batch effect optimization neural network model performs batch prediction on the multicenter pathological image feature vector data and the batch training labels to obtain the batch prediction probability value of the multicenter pathological image; the batch adversarial module based on the batch effect optimization neural network model performs backpropagation update on the batch prediction probability value of the multicenter pathological image and the batch training labels to obtain optimized multicenter pathological image data.

[0048] In this embodiment, the batch effect optimization neural network model includes a data processing module, a batch discriminator training module, and a batch adversarial module. The data processing module takes data from multiple centers as input and outputs preprocessed data such as slices. The output of the data processing module serves as the input data for the main task model. The main task model extracts feature vectors from all data, and these feature vectors serve as inputs to the batch discriminator training module and the batch adversarial module. The output of the batch discriminator training module is connected to the batch discriminator in the batch adversarial module, and the output of the batch adversarial module is used to assist the main task training to improve model performance.

[0049] The preprocessed multicenter pathological image data is input into a batch effect optimization neural network model; feature vectors are extracted through the main task model; the batch discriminator is trained using the feature vectors extracted by the main task model and the corresponding batch labels, thereby enhancing the effect of the batch discriminator module; the batch adversarial module prompts the model to generate feature vectors with reduced batch effects, thereby enhancing the consistency of data from different centers; finally, the performance of the main task model is optimized and improved through the mutual promotion effect of the batch discriminator module and the batch adversarial module.

[0050] It should be noted that the batch discriminator training module learns the deep features of multi-center pathological image data to identify the batch features of each data center, thereby enhancing the discrimination effect of the batch discriminator. The batch adversarial module and the batch discriminator training module form an adversarial effect, which enhances each other. The batch adversarial module guides the model to generate feature vectors that are less affected by batch effects through batch adversarial loss, thereby reducing the visible differences between different batches of data.

[0051] Furthermore, the batch discriminator training module selects the top k patches with the highest scores as training data. During batch discriminator validation, if the loss after three validations is less than the currently recorded minimum validation loss, the batch discriminator training is stopped. This ensures that the discriminator receives sufficient training while maintaining a certain level of training efficiency.

[0052] Furthermore, it should be noted that the data processing procedure for the batch discriminator training module is as follows:

[0053] Multicenter pathological image feature vector data and batch training labels are input into a batch discriminator training module based on a batch effect-optimized neural network model. The batch discriminator training module includes a first linear transformation layer, a first nonlinear transformation layer, a second linear transformation layer, and a second nonlinear transformation layer. Based on the first linear transformation layer of the batch discriminator training module, the feature dimension of the multicenter pathological image feature vector data is halved to obtain linearly transformed multicenter pathological image feature vector data. Based on the first nonlinear transformation layer of the batch discriminator training module, a nonlinear transformation is performed on the linearly transformed multicenter pathological image feature vector data to obtain nonlinearly transformed multicenter pathological image feature vector data. Based on the second linear transformation layer of the batch discriminator training module, a linear transformation is performed on the nonlinearly transformed multicenter pathological image feature vector data to obtain the batch prediction probability of the multicenter pathological image. Based on the second nonlinear transformation layer of the batch discriminator training module, a nonlinear transformation and confidence conversion are performed on the batch prediction probability of the multicenter pathological image and the batch training labels to obtain the batch prediction probability value of the multicenter pathological image.

[0054] Specifically, in the batch adversarial module, the batch adversarial losses obtained by each discriminator need to be summed with the main task loss using special weights. This is used to handle different batches of data and improve the model's generalization ability. The specific operations of the batch discriminator training module include combining the batch labels with the predicted labels obtained from the discriminator's inferred feature vectors, and optimizing the network using the loss function to make the network output as close as possible to the true labels. The discriminator's specific inference process is as follows: These feature vectors first undergo a linear transformation that halves the feature dimension, followed by a non-linear transformation using an activation function to enhance the model's expressive power. Subsequently, after a second linear layer, the feature dimension is further transformed into the output, representing the predicted probability of each category. Finally, an activation function is applied to convert the output into probability values, thereby generating the predicted batch probabilities.

[0055] In summary, the data processing for the batch discriminator training module can be summarized as follows:

[0056] 1) Feature dimension compression;

[0057] First, the feature dimension is halved through a linear transformation (Fully Connected Layer) to reduce computational complexity while preserving key information.

[0058] Next, activation functions (such as ReLU, Leaky ReLU, etc.) are applied to perform nonlinear transformations to enhance the model's expressive power and enable the network to learn more complex patterns.

[0059] 2) Output layer;

[0060] After passing through a second linear layer, the feature vectors are further processed to transform their dimensions into an output with the same number of categories, representing the predicted probability for each category.

[0061] Finally, an activation function (such as Softmax) is applied to convert the output into probability values, which represent the model's confidence in each class, forming the predicted batch probabilities.

[0062] Furthermore, it should be noted that the data processing procedure for the batch adversarial module is as follows:

[0063] The batch training labels are inverted to obtain inverted batch training labels. The inverted batch training labels and the batch prediction probability values ​​of multicenter pathological images are input into the batch adversarial module of the batch effect-based optimized neural network model to calculate the batch adversarial loss value of the multicenter pathological images. The batch discriminator training module is updated through backpropagation based on the batch loss value of the multicenter pathological images to obtain the updated batch discriminator training module. Based on the updated batch discriminator training module, the optimized multicenter pathological image data is output.

[0064] Specifically, the batch adversarial module operates by combining the inverted batch label with the predicted label obtained from the discriminator's inferred feature vector, and using a loss function to optimize the network so that the predicted batch corresponding to the features generated by the model deviates as much as possible from the true label.

[0065] In summary, the data processing for the batch adversarial module can be summarized as follows:

[0066] 1) Training process;

[0067] During training, a batch adversarial strategy is employed to generate high-quality features. Specifically, inverted batch labels (i.e., replacing each label in the true labels with another label) are combined with predicted labels generated from feature vectors inferred by the discriminator.

[0068] A loss function (such as cross-entropy loss) is used to evaluate the difference between the model's generated output and the true label. The loss is used to optimize the network weights, with the goal of making the output as close as possible to the true label.

[0069] 2) Optimization process;

[0070] The backpropagation algorithm is used to update network parameters using the calculated loss value, thereby improving the model's predictive ability. Through continuous iteration, the model will gradually learn better feature representations, enabling it to perform better on new data.

[0071] Finally, the batch effect optimized neural network model of this invention is evaluated by calculating the classification accuracy (ACC) and the receiver operating characteristic curve (ROC) and its area under the curve (AUC). First, the model's prediction results are compared with the true labels, and the number of correctly predicted samples is counted. Then, this number is divided by the total number of samples to obtain the classification accuracy. The specific formula is: Accuracy = Number of correctly predicted samples / Total number of samples. Further, the true positive rate (TPR) and false positive rate (FPR) are calculated under several different threshold settings. The true positive rate is defined as the number of true positives divided by the total number of actual positives, and the false positive rate is defined as the number of false positives divided by the total number of actual negatives. Next, an ROC curve is plotted with the false positive rate on the x-axis and the true positive rate on the y-axis. Finally, the area under this curve (AUC) is calculated, and this area value is used as an evaluation index of the model's classification performance. The closer the AUC value is to 1, the better the model performance. Finally, the baseline performance of the model before using the method of this invention is compared with the performance after using it, mainly focusing on the performance improvement. Specific evaluation indicators include classification accuracy and the area under the receiver operating characteristic curve (ROC) (AUC). The established evaluation criteria are: after applying the method of this invention, the model's AUC value and classification accuracy should show a significant improvement, thereby verifying the effectiveness of the method of this invention.

[0072] In summary, this embodiment of the invention extracts image features through a feature extraction module, enhances the discrimination effect of the batch discriminator by utilizing the extracted features and batch labels, and optimizes the entire network using a batch adversarial module to reduce batch effects and improve the model's adaptability and prediction accuracy to data from multiple centers. By mitigating batch effects, this embodiment of the invention can significantly improve the accuracy of pathological image analysis. Using the newly designed batch adversarial loss and batch discriminator, the model can more effectively process image data from different centers, thereby reducing errors caused by differences in scanners or operating protocols. The evaluation results are shown in Table 1.

[0073] \ Max-MILACC MAX-MILAUC PMILACC PMILAUC Baseline 0.5813±0.0523 0.7016±0.0790 0.6187±0.0601 0.7187±0.0389 Ours 0.6875±0.0541 0.7727±0.0308 0.6500±0.0559 0.7391±0.0444

[0074] Reference Figure 2 A deep learning-based batch effect optimization system for pathological images includes:

[0075] The first module 201 is used to acquire pathological sample images and perform image preprocessing to obtain multicenter pathological image data to be processed.

[0076] The second module 202 is used to introduce the batch discriminator training module and the batch adversarial module to build a batch effect optimized neural network model.

[0077] The third module 203 is used to perform batch effect optimization processing on the multicenter pathological image data to be processed based on the batch effect optimization neural network model, so as to obtain optimized multicenter pathological image data.

[0078] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0079] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for optimizing batch effects in pathological images based on deep learning, characterized in that, Includes the following steps: Acquire pathological sample images and perform image preprocessing to obtain multicenter pathological image data to be processed; A batch discriminator training module and a batch adversarial module are introduced to construct a batch effect optimization neural network model; The batch effect optimized neural network model includes a data processing module, a batch discriminator training module, and a batch adversarial module. The first output of the data processing module is connected to the first input of the batch discriminator training module, the second output of the data processing module is connected to the first input of the batch adversarial module, the output of the batch discriminator training module is connected to the second input of the batch adversarial module, and the output of the batch adversarial module is connected to the second input of the batch discriminator training module. Determine batch training labels based on the multicenter pathological image data to be processed; The multicenter pathological image data to be processed and the batch training labels are input into the batch effect optimization neural network model; The data processing module based on the batch effect optimized neural network model performs slicing and feature extraction on the multicenter pathological image data to be processed, and obtains multicenter pathological image feature vector data. Multicenter pathological image feature vector data and batch training labels are input into a batch discriminator training module based on a batch effect optimized neural network model. The batch discriminator training module includes a first linear transformation layer, a first nonlinear transformation layer, a second linear transformation layer, and a second nonlinear transformation layer. Based on the first linear transformation layer of the batch discriminator training module, the feature dimension of the multicenter pathological image feature vector data is halved to obtain the linearly transformed multicenter pathological image feature vector data. Based on the first nonlinear transformation layer of the batch discriminator training module, a nonlinear transformation is performed on the feature vector data of the multicenter pathological image after linear transformation to obtain the feature vector data of the multicenter pathological image after nonlinear transformation. Based on the second linear transformation layer of the batch discriminator training module, linear transformation processing is performed on the feature vector data of multicenter pathological images after nonlinear transformation to obtain the batch prediction probability of multicenter pathological images. Based on the second nonlinear transformation layer of the batch discriminator training module, nonlinear transformation and confidence conversion are performed on the batch prediction probability and batch training label of multicenter pathological images to obtain the batch prediction probability value of multicenter pathological images. Invert the batch training labels to obtain the inverted batch training labels. The inverted batch training labels and the batch prediction probability values ​​of multicenter pathological images are input into the batch adversarial module based on the batch effect optimization neural network model to calculate the batch adversarial loss and obtain the batch loss value of multicenter pathological images. The batch discriminator training module is updated by backpropagation based on the batch loss value of multicenter pathological images to obtain the updated batch discriminator training module. Based on the updated batch discriminator training module, optimized multicenter pathological image data is output.

2. The method for optimizing batch effects in pathological images based on deep learning according to claim 1, characterized in that, The step of acquiring pathological sample images and performing image preprocessing to obtain multicenter pathological image data to be processed specifically includes: Image data extraction and processing are performed based on pathological sample data to obtain pathological sample images; The pathological sample images are converted into digital format using a digital pathological image scanner. The pathological sample images in digital format are sequentially processed by image enhancement, threshold segmentation, and target region cutting to obtain the multicenter pathological image data to be processed.

3. The method for optimizing batch effects in pathological images based on deep learning according to claim 2, characterized in that, It also includes the use of a topk patch selection mechanism and an early stop training strategy in the batch discriminator training module to improve the data processing efficiency of the batch discriminator training module.

4. A deep learning-based batch effect optimization system for pathological images, characterized in that, Includes the following modules: The first module is used to acquire pathological sample images and perform image preprocessing to obtain multicenter pathological image data to be processed. The second module is used to introduce the batch discriminator training module and the batch adversarial module to build a batch effect optimization neural network model. The batch effect optimized neural network model includes a data processing module, a batch discriminator training module, and a batch adversarial module. The first output of the data processing module is connected to the first input of the batch discriminator training module, the second output of the data processing module is connected to the first input of the batch adversarial module, the output of the batch discriminator training module is connected to the second input of the batch adversarial module, and the output of the batch adversarial module is connected to the second input of the batch discriminator training module. The third module is used to determine batch training labels based on the multicenter pathological image data to be processed. The multicenter pathological image data to be processed and the batch training labels are input into the batch effect optimization neural network model; The data processing module based on the batch effect optimized neural network model performs slicing and feature extraction on the multicenter pathological image data to be processed, and obtains multicenter pathological image feature vector data. Multicenter pathological image feature vector data and batch training labels are input into a batch discriminator training module based on a batch effect optimized neural network model. The batch discriminator training module includes a first linear transformation layer, a first nonlinear transformation layer, a second linear transformation layer, and a second nonlinear transformation layer. Based on the first linear transformation layer of the batch discriminator training module, the feature dimension of the multicenter pathological image feature vector data is halved to obtain the linearly transformed multicenter pathological image feature vector data. Based on the first nonlinear transformation layer of the batch discriminator training module, a nonlinear transformation is performed on the feature vector data of the multicenter pathological image after linear transformation to obtain the feature vector data of the multicenter pathological image after nonlinear transformation. Based on the second linear transformation layer of the batch discriminator training module, linear transformation processing is performed on the feature vector data of multicenter pathological images after nonlinear transformation to obtain the batch prediction probability of multicenter pathological images. Based on the second nonlinear transformation layer of the batch discriminator training module, nonlinear transformation and confidence conversion are performed on the batch prediction probability and batch training label of multicenter pathological images to obtain the batch prediction probability value of multicenter pathological images. Invert the batch training labels to obtain the inverted batch training labels. The inverted batch training labels and the batch prediction probability values ​​of multicenter pathological images are input into the batch adversarial module based on the batch effect optimization neural network model to calculate the batch adversarial loss and obtain the batch loss value of multicenter pathological images. The batch discriminator training module is updated by backpropagation based on the batch loss value of multicenter pathological images to obtain the updated batch discriminator training module. Based on the updated batch discriminator training module, optimized multicenter pathological image data is output.

Images