Pathology intelligent quality control management system based on big data

Abstract

The invention discloses an intelligent pathological quality control management system based on big data. The system comprises a multi-source pathological data collection module, a data optimization module, a pathological image quality control detection module, a pathological process abnormity tracing module and an intelligent quality control management module. The invention relates to the technical field of medical pathology data processing, in particular to an intelligent pathology quality control management system based on big data, which innovatively combines pathology image quality control detection and pathology process abnormity tracing, shortens the quality control and problem positioning time, and accelerates the pathology quality control process; wavelet transform is introduced to perform pathological image multi-source information processing, and a cavity volume accumulation mechanism is combined to expand a receptive field, so that the accuracy and performance of pathological defect detection are effectively improved; a pathological salient region quality control loss function is proposed, and the discrimination ability and the training effect of the model are improved; and an optimization algorithm is improved by introducing a worst individual position iterative correction strategy, so that the performance of the tracing model is improved, and intelligent management of the pathological quality control system is realized.

Description

Technical Field

[0001] This invention relates to the field of medical pathology data processing technology, specifically to a pathology intelligent quality control management system based on big data. Background Technology

[0002] The big data-based intelligent pathology quality control management system refers to a system that combines big data technology with medical information technology. Through the collection and integration of multi-source data, it stores, processes, and models pathology-related data, and uses machine learning and process analysis methods to achieve quality control of the entire pathology process. This system can provide intelligent early warning and visual management, and promote the automation, standardization, and intelligent development of pathology quality control.

[0003] However, traditional pathology quality control management systems suffer from several problems. First, they employ a step-by-step quality control approach, leading to cumbersome and inefficient processes. Second, existing pathology image detection models struggle to accurately distinguish defect boundaries, lack the ability to detect large-scale defects, and tend to overlook key defect features, resulting in low detection accuracy. Third, the loss functions used in training these models fail to adequately consider the importance and salience of different defect regions in the pathology image, further contributing to low defect detection accuracy. Fourth, existing models for tracing pathology workflow anomalies often suffer from unreasonable hyperparameter settings, slow convergence during optimization, and a tendency to get trapped in local optima, making it difficult to find the globally optimal hyperparameter combination and resulting in insufficient accuracy in the model's output. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a big data-based intelligent pathology quality control management system. Addressing the technical problem of traditional pathology quality control systems employing a step-by-step quality control approach, resulting in cumbersome and inefficient processes, this solution innovatively combines pathology image quality control detection with pathology workflow anomaly tracing. Based on reverse tracing of pathology image issues, quality control is applied only to relevant workflow steps, significantly improving quality control efficiency, shortening quality control and problem location time, accelerating the pathology quality control process, and accurately locating the specific steps and causes of anomalies. This enhances the accuracy of pathology quality control problem location and achieves automated, standardized, and intelligent management of pathology quality control. Furthermore, it addresses the limitations of existing systems applicable to pathology images... This solution addresses the technical challenges of existing pathological image detection models, such as difficulty in accurately distinguishing defect boundaries, insufficient detection capability for large-scale defect areas, and the tendency to overlook key defect features, leading to low detection accuracy. It innovatively introduces wavelet transform for multi-source information processing of pathological images, improving the segmentation accuracy of pathological defect boundaries; combines dilated convolution aggregation mechanism to expand the receptive field, enhancing the model's ability to recognize large-scale defect areas; and utilizes channel and spatial attention mechanisms to highlight key defect areas, increasing attention to significant pathological defect regions. This effectively improves the accuracy and performance of pathological defect detection, achieving intelligent quality control of digital pathological images. Furthermore, it addresses the limitations of existing loss functions used in training pathological image detection models. To address the technical problem of low accuracy in pathological image defect detection due to insufficient consideration of the importance and salience of different defect regions in pathological images, this solution innovatively proposes a pathological salience region quality control loss function. This function weights and sums the cross-entropy loss term and the salience penalty term, introducing a salience guidance mechanism to enhance the model's focus on key defect regions. By gradually increasing the salience weight during training, the model is guided to focus more on salience defect regions in pathological images, thereby effectively improving the model's discrimination ability and training effect, increasing the accuracy of pathological image defect boundary recognition, enhancing the model's response to key regions, and significantly improving the overall reliability of pathological image quality control detection. This addresses the shortcomings of existing models applicable to pathological workflow anomaly tracing. The technical problem of insufficient accuracy in model output results due to unreasonable hyperparameter settings, slow convergence speed, and susceptibility to local optima during hyperparameter optimization, which prevents the finding of the globally optimal hyperparameter combination, is addressed by this innovative solution. This solution introduces an iterative correction strategy for the worst-fit individual position in the optimization algorithm. Through a periodic oscillation guidance mechanism, the position of the worst-fit individual in the current iteration is adjusted, allowing it to escape local extreme regions. This increases the diversity of solution space exploration, thereby enhancing the search process's ability to find the globally optimal solution. This improves the global search capability and convergence speed of the optimization algorithm, increases the accuracy of the output results of the pathological process anomaly tracing model, strengthens the ability to locate abnormal links in the pathological process, and realizes intelligent and refined management of the pathological quality control system.

[0005] The technical solution adopted by the present invention is as follows: The pathological intelligent quality control management system based on big data provided by the present invention includes a multi-source pathological data collection module, a data optimization module, a pathological image quality control detection module, a pathological process abnormality tracing module, and an intelligent quality control management module;

[0006] The multi-source pathological data collection module specifically obtains raw pathological quality control management data through data acquisition operations.

[0007] The data optimization module specifically performs pathological image data preprocessing, pathological process management data preprocessing, and multimodal data alignment to obtain optimized pathological quality control management data.

[0008] The pathological image quality control and detection module is used to perform automated quality detection and problem identification on digital pathological images. Specifically, it first establishes a pathological image detection model, uses historical pathological image detection data as training data and adopts a pathological salient region quality control loss function as a supervision target to train the pathological image detection model, and finally inputs real-time pathological image detection data into the trained model to obtain the pathological image defect quality control detection results.

[0009] The pathological process anomaly tracing module is used to trace the abnormal process links corresponding to pathological image quality problems. Specifically, it constructs a pathological process anomaly tracing model based on a bidirectional long short-term memory neural network, trains the model, obtains the hyperparameter optimization combination of the pathological process anomaly tracing model by improving the search algorithm, and obtains the pathological process anomaly tracing model with the best performance. The real-time pathological process management data corresponding to the pathological image and the pathological image defect quality control detection results are input into the model to obtain the pathological process anomaly identification results.

[0010] The intelligent quality control management module is used to realize intelligent quality control management of the pathological process. Specifically, it monitors and controls the quality of the pathological process in real time by combining the quality control detection results of pathological image defects and the identification results of pathological process anomalies.

[0011] Furthermore, the multi-source pathological data collection module specifically obtains raw pathological quality control management data by acquiring data from the pathology laboratory information system and the hospital information system. This raw pathological quality control management data includes historical pathological image detection data, historical pathological process management data, real-time pathological image detection data, and real-time pathological process management data. Both historical and real-time pathological image detection data include digital pathological images. The historical pathological image detection data also includes historical pathological image detection results. Both historical and real-time pathological process management data include pathological equipment operation log data, pathological operation process record data, and pathological operation environment data. The historical pathological process management data also includes operation defect record data.

[0012] Furthermore, the data optimization module is used to optimize the raw data of pathology quality control management, specifically by performing pathology image data preprocessing, pathology workflow management data preprocessing, and multimodal data alignment to obtain optimized pathology quality control management data, including the following steps:

[0013] Preprocessing of pathological image data includes image cleaning, image standardization, image quality enhancement, and image annotation to obtain optimized data for pathological image detection.

[0014] Preprocessing of pathology workflow management data includes data cleaning, data standardization, and data coding to obtain optimized pathology workflow data.

[0015] Multimodal data alignment is used to achieve a one-to-one correspondence between the image quality of the same pathological sample and the data of the process steps. Specifically, it is achieved by aligning the pathological image data with the corresponding process management data according to the sample number and timestamp.

[0016] Furthermore, the pathological image quality control and detection module is used to perform automated quality detection and problem identification on pathological digital images, specifically including the following steps:

[0017] The establishment of a pathological image detection model includes the following steps:

[0018] The multi-source information processing of pathological images involves first performing a two-dimensional discrete wavelet transform on the digital pathological image to decompose it into low-frequency components and high-frequency components in three directions. Then, convolution, normalization, and nonlinear activation are applied to the four types of components to obtain the corresponding sub-band enhancement features. Next, the sub-band enhancement features are reconstructed from the frequency domain to the spatial domain using inverse wavelet transform to obtain the image spatial reconstruction features. Finally, the image spatial reconstruction features are concatenated with the image shallow convolution features in the channel dimension, and the number of channels is compressed through convolution to obtain the final shallow enhancement features.

[0019] The deep feature extraction of pathological images involves inputting shallow enhancement features into a ResNet50-based neural network, performing initial convolution and layer-by-layer convolution and downsampling in four residual stages to generate a multi-scale deep feature set, and then aligning and compressing the multi-scale feature set through 1×1 convolution to obtain the deep features of the pathological image.

[0020] The global pathological information aggregation specifically involves inputting deep features of the pathological image into a dilated convolutional aggregation layer, generating the convolutional outputs of each branch, and adding them element-wise with the deep features of the pathological image to obtain global context-enhanced features; the dilated convolutional aggregation layer consists of... It consists of several parallel convolutional branches, each with a different dilation rate; the formula used is as follows:

[0021] ;

[0022] In the formula, This represents a global context-enhanced feature. Indicates deep features in pathological images. This indicates an element-wise addition operation. Indicates the number of convolution branches. Let represent the dilation rate used in the i-th branch, and d represent the dilation coefficient of the convolution. This indicates that the kernel size is 3×3 and the dilation rate is... The dilated convolution operation;

[0023] The enhancement of pathologically significant defect regions involves taking global context enhancement features as input, first applying a channel attention mechanism to obtain channel attention weighted features, then applying a spatial attention mechanism to the channel attention weighted features to obtain a spatial weight map, and finally weighting the channel attention weighted features in the spatial dimension to obtain significant enhancement features.

[0024] The defect detection results are output by inputting the saliency enhancement features into the convolution classifier head, calculating the probability distribution of each pathological image defect category through the Softmax activation function, and selecting the category corresponding to the highest probability as the pathological image defect detection result.

[0025] The training of the pathological image detection model involves using historical pathological image detection data as training data, employing a pathological salient region quality control loss function as the supervised training objective function, and training the pathological image detection model through gradient descent iterative optimization to finally obtain the trained pathological image detection model.

[0026] The pathological salient region quality control loss function is used to introduce salient region constraints during the supervised training of the pathological image detection model, making the model pay more attention to salient defect regions in pathological slices during optimization. Specifically, it is obtained by weighting and summing the cross-entropy loss term and the salientity penalty term according to linear dynamic weight parameters; the formula used is as follows:

[0027] ;

[0028] ;

[0029] ;

[0030] In the formula, This represents the significance penalty weight corresponding to the e-th training epoch. This represents the initial weight value. This represents the maximum weight value. Indicates the current training cycle. Indicates the total number of training cycles. The value represents the significance penalty term, B represents the number of pathological digital images used in the training, and b represents the index of the pathological digital images. This indicates the number of features in this image. The saliency score of the b-th pathological digital image on feature j is calculated using the SHAP method. This represents the quality control loss function value for pathologically significant regions. This represents the value of the cross-entropy loss function;

[0031] Pathological image defect quality control specifically involves inputting real-time pathological image detection data into a trained pathological image detection model and outputting pathological image defect quality control detection results.

[0032] Furthermore, the pathological process anomaly tracing module specifically includes the following steps:

[0033] A pathological process anomaly tracing model was constructed and trained. Specifically, a pathological process anomaly tracing model was established based on a bidirectional long short-term memory neural network. The historical pathological process management data corresponding to the pathological images and the detection results of historical pathological images were used as training data for the model. The pathological process anomaly tracing model was trained to obtain the trained pathological process anomaly tracing model.

[0034] Model hyperparameter optimization includes the following steps:

[0035] Initialize the search population by encoding the hyperparameters of the trained pathological process anomaly tracing model into search individual position vectors, and generate N search individual position vectors through a random initialization method. Each individual encoding represents a candidate combination of pathological process anomaly tracing model hyperparameters to obtain the initial search population.

[0036] Population role assignment, specifically, calculating the fitness value of the searched individual in the population. The performance of the pathological process anomaly tracing model based on individual location is used as the fitness value of the individual. Individuals are ranked from best to worst according to their fitness value. The top 20% of search individuals are classified as producers, and the remaining 80% of search individuals are classified as foragers.

[0037] Producer position updates, specifically based on danger signals. and safety threshold Determine the location and update it accordingly;

[0038] The worst-performing individual's position is iteratively corrected by employing a periodic oscillation-guided strategy to adjust the position of the worst-performing individual in the population; the formula used is as follows:

[0039] ;

[0040] In the formula, This indicates the position of the worst-performing individual in the current iteration after correction. This indicates the position of the worst-performing individual in the current iteration. and Both represent random numbers in the range [0,1]. express Random numbers within a range This indicates the optimal individual position in the current iteration. This indicates the position of the worst-case individual globally. This indicates the maximum number of iterations. The factor representing the control of periodic oscillations is a constant, and its value ranges from [value missing]. ;

[0041] Forager location updates are specifically based on individual indexes and population size.

[0042] Danger detectors update their locations by randomly selecting 10% of individuals in the population as danger detectors.

[0043] The best and worst individuals are updated by recalculating the fitness values ​​of all searched individuals and determining the new global best and worst individual positions based on the fitness results.

[0044] The search terminates, specifically when the fitness value of the searched individual is... When the fitness threshold is exceeded or the maximum number of iterations is reached, the search is terminated and the globally optimal individual position is obtained; the globally optimal individual position specifically refers to the optimized combination of hyperparameters.

[0045] The model hyperparameter update specifically involves adjusting the hyperparameters of the pathological process anomaly tracing model based on the optimized combination of the hyperparameters to obtain the optimal-performing pathological process anomaly tracing model.

[0046] The pathological process anomaly identification specifically involves inputting the real-time pathological process management data corresponding to the pathological image and the pathological image defect quality control detection results into the optimal-performing pathological process anomaly tracing model to obtain the pathological process anomaly identification result.

[0047] Furthermore, the intelligent quality control management module is used to realize intelligent quality control management of the pathological process. Specifically, by combining the quality control detection results of pathological image defects and the identification results of pathological process anomalies, it automatically identifies the correlation between pathological image quality problems and front-end pathological process anomalies, monitors and controls the quality of the pathological process in real time, thereby realizing comprehensive quality control and intelligent management of the pathological process.

[0048] The beneficial effects achieved by the present invention using the above solution are as follows:

[0049] (1) In view of the technical problem that the traditional pathological quality control management system adopts a step-by-step quality control method, which leads to the cumbersome and inefficient pathological quality control process, this solution innovatively combines pathological image quality control detection with pathological process anomaly tracing. Based on the reverse tracing of pathological image problems, only the relevant process links are quality controlled, thereby significantly improving the quality control efficiency, shortening the quality control and problem location time, accelerating the pathological quality control process, accurately locating the specific link and cause of the anomaly, improving the accuracy of pathological quality control problem location, and realizing the level of automated, standardized and intelligent management of pathological quality control.

[0050] (2) In view of the technical problems in existing pathological image detection models, such as difficulty in accurately distinguishing the boundaries of defect areas, insufficient detection capability for large-scale defect areas, and easy neglect of key defect features, resulting in low accuracy of pathological image detection, this solution innovatively improves the segmentation accuracy of pathological defect boundaries by introducing wavelet transform for multi-source information processing of pathological images; expands the receptive field by combining dilated convolution aggregation mechanism to enhance the model's ability to identify large-scale defect areas; and highlights key defect areas by using channel and spatial attention mechanisms to enhance attention to significant areas of pathological defects; effectively improves the accuracy and performance of pathological defect detection and realizes intelligent quality control of digital pathological images.

[0051] (3) In view of the technical problem that the loss function used in the training of existing pathological image detection models fails to fully consider the importance and salience of different defect regions in pathological images, resulting in low accuracy of pathological image defect detection, this scheme innovatively proposes a pathological salience region quality control loss function. The cross-entropy loss term and the salience penalty term are weighted and summed to introduce a salience guidance mechanism to enhance the model’s attention to key defect regions. By gradually increasing the salience weight during the training process, the model is guided to focus more on the salience defect regions in pathological images, thereby effectively improving the model’s discrimination ability and training effect, improving the recognition accuracy of pathological image defect boundaries, enhancing the model’s response ability to key regions, and significantly improving the overall reliability of pathological image quality control detection.

[0052] (4) In view of the technical problems of the existing pathological process abnormality tracing model, which has unreasonable hyperparameter settings, slow convergence speed and easy to get trapped in local optima during the hyperparameter optimization process, and cannot find the global optimal hyperparameter combination, resulting in insufficient accuracy of the model output results, this solution innovatively introduces the worst individual position iterative correction strategy in the optimization algorithm. Through the periodic oscillation guidance mechanism, the position of the individual with the worst fitness in the current iteration is adjusted so that it jumps out of the local extreme value region, increases the diversity of solution space exploration, thereby improving the search ability of the search process to find the global optimal solution, improving the global search ability and convergence speed of the optimization algorithm, and improving the accuracy of the output results of the pathological process abnormality tracing model, enhancing the ability to locate abnormal links in the pathological process, and realizing the intelligent and refined management of the pathological quality control system. Attached Figure Description

[0053] Figure 1 A schematic diagram of the modules of the big data-based intelligent pathology quality control management system provided by the present invention;

[0054] Figure 2 This is a flowchart illustrating the pathology image quality control and detection module.

[0055] Figure 3 A flowchart illustrating the process of establishing a pathological image detection model in the pathological image quality control and detection module;

[0056] Figure 4 This is a flowchart illustrating the abnormality tracing module in the pathology workflow.

[0057] Figure 5 A flowchart illustrating the hyperparameter optimization of the mid-model in the abnormal traceability module of the pathology process;

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0059] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0060] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0061] Example 1, see Figure 1 The pathology intelligent quality control management system based on big data provided by the present invention includes a multi-source pathology data collection module, a data optimization module, a pathology image quality control detection module, a pathology process anomaly tracing module, and an intelligent quality control management module.

[0062] The multi-source pathological data collection module specifically obtains the original pathological quality control management data through data acquisition operations and sends the data to the data optimization module.

[0063] The data optimization module receives data sent by the multi-source pathological data collection module, specifically performs pathological image data preprocessing, pathological process management data preprocessing, and multimodal data alignment to obtain pathological quality control management optimization data, and sends the data to the pathological image quality control detection module and the pathological process anomaly tracing module.

[0064] The pathological image quality control and detection module receives data sent by the data optimization module and is used to perform automated quality detection and problem identification on pathological digital images. Specifically, it first establishes a pathological image detection model, uses historical pathological image detection data as training data and adopts the pathological significant region quality control loss function as the supervision target to train the pathological image detection model. Finally, it inputs real-time pathological image detection data into the trained model to obtain the pathological image defect quality control detection results, and sends the data to the pathological process anomaly tracing module and the intelligent quality control management module.

[0065] The pathological process anomaly tracing module receives data sent by the data optimization module and the pathological image quality control detection module, and is used to trace the abnormal process links corresponding to pathological image quality problems. Specifically, it constructs a pathological process anomaly tracing model based on a bidirectional long short-term memory neural network, trains the model, obtains the hyperparameter optimization combination of the pathological process anomaly tracing model by improving the search algorithm, and obtains the pathological process anomaly tracing model with the best performance. The real-time pathological process management data corresponding to the pathological image and the pathological image defect quality control detection results are input into the model to obtain the pathological process anomaly identification results, and the data is sent to the intelligent quality control management module.

[0066] The intelligent quality control management module receives data from the pathological process anomaly tracing module and the pathological image quality control detection module to realize intelligent quality control management of the pathological process. Specifically, it monitors and controls the quality of the pathological process in real time by combining the pathological image defect quality control detection results and the pathological process anomaly identification results.

[0067] By performing the above operations, this solution addresses the technical problem in traditional pathology quality control management systems where a step-by-step quality control approach leads to cumbersome and inefficient processes. It innovatively combines pathology image quality control detection with pathology workflow anomaly tracing. Based on reverse tracing of pathology image issues, quality control is applied only to relevant workflow steps, significantly improving quality control efficiency, shortening quality control and problem location time, accelerating the pathology quality control process, accurately locating the specific steps and causes of anomalies, enhancing the accuracy of pathology quality control problem location, and achieving automated, standardized, and intelligent management of pathology quality control.

[0068] Example 2, see Figure 1 This embodiment is based on the above embodiment. Specifically, the multi-source pathological data collection module obtains raw pathological quality control management data by acquiring data from the pathology laboratory information system and the hospital information system. The raw pathological quality control management data includes historical pathological image detection data, historical pathological process management data, real-time pathological image detection data, and real-time pathological process management data. Both the historical and real-time pathological image detection data include digital pathological images. The historical pathological image detection data also includes historical pathological image detection results. Both the historical and real-time pathological process management data include pathological equipment operation log data, pathological operation process record data, and pathological operation environment data. According to the above, the historical pathology process management data also includes operational defect record data; the pathology equipment operation log data includes equipment type, equipment operation parameters, and equipment operation status; the pathology operation process record data refers to the operation record data of operators in the pathology sample preparation process, including operation step records, operation time, operator information, and pathology sample information; the pathology operation environment data refers to data related to the laboratory environment and external conditions, including laboratory environment parameters, pathology sample storage conditions, and laboratory workload data; the pathology sample preparation process includes pathology sample fixation, pathology sample dehydration, pathology sample embedding, pathology sample sectioning, pathology sample staining, pathology mounting, and pathology image generation.

[0069] Example 3, see Figure 1 This embodiment is based on the above embodiment. The data optimization module is used to optimize the raw data of pathology quality control management. Specifically, it performs pathology image data preprocessing, pathology process management data preprocessing, and multimodal data alignment to obtain optimized pathology quality control management data. The steps include:

[0070] Pathological image data preprocessing is used to improve the usability and consistency of pathological images. Specifically, it includes image cleaning, image standardization, image quality enhancement, and image annotation to obtain optimized data for pathological image detection.

[0071] The image cleaning is used to remove invalid or low-quality areas in pathological images. Specifically, it removes blurred areas, scanning artifacts and non-tissue areas in the slices through edge detection, noise filtering algorithms and blank area cropping methods.

[0072] The image standardization process is used to ensure the consistency of color and resolution of images generated in different batches and from different devices; specifically, it standardizes the images through color normalization, resolution unification algorithms, and slice size scaling methods.

[0073] The image quality enhancement is used to improve the visibility of pathological images, specifically by performing random geometric transformations and random brightness and contrast adjustments.

[0074] The image annotation is used to provide training labels for subsequent pathological image quality detection and diagnostic models. Specifically, it involves semi-automatic annotation of out-of-focus areas, bubble areas, abnormally stained areas, tissue missing areas, and scanning artifact areas in pathological sections based on historical pathological image detection results.

[0075] Pathology workflow management data preprocessing is used to improve the structure and analyzability of pathology workflow-related data. Specifically, it includes data cleaning, data standardization, and data coding to obtain optimized pathology workflow data.

[0076] The data cleaning is used to remove redundant, missing and abnormal records from the pathology process data. Specifically, it involves cleaning the pathology process management data using the mean-filling method and the Z-Score algorithm.

[0077] The data standardization is used to unify the data format and numerical scale generated by different data sources and devices. Through the min-max normalization method, process data is converted into a standard format with strong comparability.

[0078] The data encoding process is used to transform unstructured process text and logs into an input format acceptable to the model. Specifically, it involves extracting and encoding keywords from operation records using natural language processing methods, and converting structured information into numerical vectors using one-hot encoding.

[0079] Multimodal data alignment is used to achieve a one-to-one correspondence between the image quality of the same pathological sample and the data of the process steps. Specifically, it is achieved by aligning the pathological image data with the corresponding process management data according to the sample number and timestamp.

[0080] Example 4, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. The pathological image quality control and detection module is used to perform automated quality detection and problem identification on pathological digital images, specifically including the following steps:

[0081] The establishment of a pathological image detection model includes the following steps:

[0082] Multi-source information processing of pathological images is used to preprocess the input original pathological slice images with multi-source features. Specifically, firstly, a two-dimensional discrete wavelet transform is performed on the digital pathological image to decompose the image into low-frequency components and high-frequency components in three directions. Then, convolution operations, normalization, and nonlinear activation are applied to the four types of components to obtain the corresponding sub-band enhancement features. Next, the sub-band enhancement features are reconstructed from the frequency domain to the spatial domain using inverse wavelet transform to obtain the image spatial reconstruction features. Then, the image spatial reconstruction features are concatenated with the image shallow convolution features in the channel dimension, and the number of channels is compressed by convolution to obtain the final shallow enhancement features.

[0083] The high-frequency components in the three directions are respectively the high-frequency components in the horizontal direction. Vertical high frequency components High-frequency components in the diagonal edge direction ;

[0084] The image shallow convolution feature is specifically obtained by applying convolution, normalization and activation operations to the pathological digital image;

[0085] The formula used is as follows:

[0086] ;

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] In the formula, Indicates low-frequency components. This represents the two-dimensional discrete wavelet transform function. This represents the j-th component. This represents the enhanced feature of the j-th subband. Indicates low frequency. Indicates high frequency in the horizontal direction. Indicates high frequency in the vertical direction. Indicates high frequency in the direction of the diagonal edge. This indicates a normalization operation. Represents the ReLU activation function. Representing pathological digital images, Describes the inverse wavelet transform function. This represents a 3×3 convolution operation. This represents a 1×1 convolution operation. Represents image spatial reconstruction features, This indicates the low-frequency component subband enhancement feature. This indicates a channel-level concatenation operation. This indicates the high-frequency component subband enhancement characteristics in the horizontal direction. This indicates the vertical high-frequency component subband enhancement feature. This indicates the high-frequency component subband enhancement feature along the diagonal edge direction. Represents the shallow convolutional features of an image. Indicates shallow enhancement features;

[0092] Deep feature extraction of pathological images is used to perform multi-layer convolutional modeling on shallow enhancement features to obtain discriminative high-level semantic features. Specifically, the shallow enhancement features are input into a ResNet50-based neural network, and after initial convolution and layer-by-layer convolution and downsampling in four residual stages, a multi-scale deep feature set is generated. The multi-scale feature set is then channel-aligned and compressed using 1×1 convolution to obtain the deep features of the pathological image.

[0093] Global pathological information aggregation is used to expand the receptive field of the model while maintaining image resolution, thereby obtaining global semantic context information. Specifically, deep features of the pathological image are input into a dilated convolution aggregation layer to generate the convolution output of each branch. The output is then added element-wise with the deep features of the pathological image to obtain global context enhancement features. The dilated convolution aggregation layer consists of... It consists of several parallel convolutional branches, each with a different dilation rate; the formula used is as follows:

[0094] ;

[0095] In the formula, This represents a global context-enhanced feature. Indicates deep features in pathological images. This indicates an element-wise addition operation. Indicates the number of convolution branches. Let represent the dilation rate used in the i-th branch, and d represent the dilation coefficient of the convolution. This indicates that the kernel size is 3×3 and the dilation rate is... The dilated convolution operation;

[0096] Enhancement of Pathologically Significant Defect Regions: This method aims to enhance the expression of pathologically defective regions. Specifically, it takes global context enhancement features as input, first applies a channel attention mechanism to obtain channel attention weighted features, then applies a spatial attention mechanism to the channel attention weighted features to obtain a spatial weight map, and finally weights the channel attention weighted features in the spatial dimension to obtain significant enhancement features.

[0097] The channel attention mechanism specifically involves first generating a channel weight vector through global average pooling, followed by two layers of fully connected mapping and Sigmoid activation, and then recalibrating the features of each channel to generate channel-weighted features.

[0098] The spatial attention mechanism specifically involves performing average pooling and max pooling operations on the channel-weighted features to generate two single-channel feature maps, which are then concatenated in the channel dimension and input into a convolutional layer, activated by Sigmoid to generate a spatial weight map.

[0099] The formula used is as follows:

[0100] ;

[0101] ;

[0102] ;

[0103] In the formula, This indicates a global average pooling operation. This represents the weight matrix of the first fully connected layer. This represents the weight matrix of the second fully connected layer. This indicates an element-wise multiplication operation. This represents the channel attention weighting feature. Represents a spatial weighted graph. This represents a 7×7 convolution operation. Indicates a feature with enhanced saliency. This indicates a max pooling operation. This indicates the average pooling operation;

[0104] The defect detection results are output to support defect type identification in pathological quality control. Specifically, the saliency enhancement features are input into the convolution classifier head, the probability distribution of each pathological image defect category is calculated through the Softmax activation function, and the category corresponding to the highest probability is selected as the pathological image defect detection result.

[0105] The training of the pathological image detection model involves using historical pathological image detection data as training data, employing a pathological salient region quality control loss function as the supervised training objective function, and training the pathological image detection model through gradient descent iterative optimization to finally obtain the trained pathological image detection model.

[0106] The pathological salient region quality control loss function is used to introduce salient region constraints during the supervised training of the pathological image detection model, making the model pay more attention to salient defect regions in pathological slices during optimization. Specifically, it is obtained by weighting and summing the cross-entropy loss term and the salientity penalty term according to linear dynamic weight parameters; the formula used is as follows:

[0107] ;

[0108] ;

[0109] ;

[0110] In the formula, This represents the significance penalty weight corresponding to the e-th training epoch. This represents the initial weight value. This represents the maximum weight value. Indicates the current training cycle. Indicates the total number of training cycles. The value represents the significance penalty term, B represents the number of pathological digital images used in the training, and b represents the index of the pathological digital images. This indicates the number of features in this image. The saliency score of the b-th pathological digital image on feature j is calculated using the SHAP method. This represents the quality control loss function value for pathologically significant regions. This represents the value of the cross-entropy loss function;

[0111] Pathological image defect quality control specifically involves inputting real-time pathological image detection data into a trained pathological image detection model and outputting pathological image defect quality control detection results.

[0112] By performing the above operations, this solution addresses the technical problems in existing pathological image detection models, such as difficulty in accurately distinguishing defect region boundaries, insufficient detection capability for large-scale defect regions, and easy neglect of key defect features, which lead to low accuracy in pathological image detection. This solution innovatively introduces wavelet transform for multi-source information processing of pathological images, improving the segmentation accuracy of pathological defect boundaries; combines dilated convolution aggregation mechanism to expand the receptive field, enhancing the model's ability to recognize large-scale defect regions; and utilizes channel and spatial attention mechanisms to highlight key defect regions, increasing attention to significant pathological defect areas. This effectively improves the accuracy and performance of pathological defect detection, achieving intelligent quality control of digital pathological images.

[0113] To address the technical problem that existing loss functions used in training pathological image detection models fail to adequately consider the importance and salience of different defect regions in pathological images, resulting in low accuracy in pathological image defect detection, this solution innovatively proposes a pathological salience region quality control loss function. This function weights and sums the cross-entropy loss term and the salience penalty term, introducing a salience guidance mechanism to enhance the model's focus on key defect regions. By gradually increasing the salience weight during training, the model is guided to focus more on salience defect regions in pathological images, thereby effectively improving the model's discrimination ability and training effect, increasing the accuracy of pathological image defect boundary recognition, enhancing the model's response to key regions, and significantly improving the overall reliability of pathological image quality control detection.

[0114] Example 5, see Figure 1 , Figure 4 and Figure 5 This embodiment is based on the above embodiment, and the pathological process abnormality tracing module specifically includes the following steps:

[0115] A pathological process anomaly tracing model was constructed and trained. Specifically, a pathological process anomaly tracing model was established based on a bidirectional long short-term memory neural network. The historical pathological process management data corresponding to the pathological images and the detection results of historical pathological images were used as training data for the model. The pathological process anomaly tracing model was trained to obtain the trained pathological process anomaly tracing model.

[0116] Model hyperparameter optimization specifically involves obtaining an optimized combination of hyperparameters for the pathological workflow anomaly tracing model through an improved search algorithm; this includes the following steps:

[0117] Initialize the search population by encoding the hyperparameters of the trained pathological process anomaly tracing model into search individual position vectors, and generate N search individual position vectors through a random initialization method. Each individual encoding represents a candidate combination of pathological process anomaly tracing model hyperparameters to obtain the initial search population.

[0118] Population role assignment, specifically calculating the fitness values of search individuals in the population , taking the performance of the pathological process anomaly tracing model established based on the individual's position as the fitness value of the individual, and sorting the individuals from the best to the worst according to the fitness value. The top 20% of the search individuals are classified as producers, and the remaining 80% of the search individuals are foragers;

[0119] Producer position update, specifically judging according to the danger signal and safety threshold to perform position update; the formula used is as follows:

[0120] ;

[0121] In the formula, represents the position of the i-th individual in the d-th dimension of the (t + 1)-th generation population, represents the position of the i-th individual in the d-th dimension of the t-th generation population, represents the corresponding position of the best fitness in the current iteration. R2 < ST indicates being in a safe position, R2 ≥ ST indicates there is an enemy nearby and a different area needs to be changed. Q represents a random number following a normal distribution, represents the search individual index, represents the maximum number of iterations, L represents a vector with all elements being 1, represents the parameter controlling the exponential convergence speed, and its value range is ;

[0122] Iterative correction of the position of the worst individual, specifically using a periodic oscillation guiding strategy to correct the position of the worst individual in the population; the formula used is as follows:

[0123] ;

[0124] In the formula, represents the position of the worst individual in the current iteration after correction, represents the position of the worst individual in the current iteration, and both represent random numbers within the range of [0, 1], represents a random number within the range of represents the position of the best individual in the current iteration, represents the position of the globally worst individual, represents the factor controlling the periodic oscillation guidance, which is a constant, and its value range is ;

[0125] Forager position update, specifically performing position update according to the individual index and population size; the formula used is as follows:

[0126] ;

[0127] In the formula, n represents the number of individuals. Let A represent the optimal individual position in the (t+1)th iteration. + express A pseudo-inverse matrix with elements randomly assigned 1 or -1;

[0128] Danger detectors update their locations by randomly selecting 10% of individuals in the population as danger detectors; the formula used is as follows:

[0129] ;

[0130] In the formula, Let represent a standard normal random number with a mean of 0 and a variance of 1, and K represent a uniformly distributed random number in the interval [−1, 1]. This represents the fitness value of the globally optimal individual. This represents the fitness value of the worst individual in the current iteration after correction. Represents a minimal constant;

[0131] The best and worst individuals are updated by recalculating the fitness values ​​of all searched individuals and determining the new global best and worst individual positions based on the fitness values.

[0132] The search terminates, specifically when the fitness value of the searched individual is... When the fitness threshold is exceeded or the maximum number of iterations is reached, the search is terminated and the globally optimal individual position is obtained; the globally optimal individual position specifically refers to the optimized combination of hyperparameters.

[0133] The model hyperparameter update specifically involves adjusting the hyperparameters of the pathological process anomaly tracing model based on the optimized combination of the hyperparameters to obtain the optimal-performing pathological process anomaly tracing model.

[0134] The pathological process anomaly identification specifically involves inputting the real-time pathological process management data corresponding to the pathological image and the pathological image defect quality control detection result into the optimal-performing pathological process anomaly tracing model to obtain the pathological process anomaly identification result.

[0135] The abnormal identification results of the pathological process specifically refer to the abnormalities that occur in the pathological sample preparation process before the generation of pathological images.

[0136] By performing the above operations, this solution addresses the technical problems of existing models for tracing pathological workflow anomalies, such as unreasonable hyperparameter settings, slow convergence speed during hyperparameter optimization, susceptibility to local optima, and inability to find the globally optimal hyperparameter combination, leading to insufficient accuracy of model output results. This solution innovatively introduces an iterative correction strategy for the worst-fit individual position into the optimization algorithm. Through a periodic oscillation guidance mechanism, the position of the worst-fit individual in the current iteration is adjusted, allowing it to escape local extreme regions. This increases the diversity of solution space exploration, thereby enhancing the search process's ability to find the globally optimal solution. This improves the global search capability and convergence speed of the optimization algorithm, increases the accuracy of the output results of the pathological workflow anomaly tracing model, strengthens the ability to locate abnormal links in the pathological workflow, and realizes intelligent and refined management of the pathological quality control system.

[0137] Example 6, see Figure 1 This embodiment is based on the above embodiment. The intelligent quality control management module is used to realize intelligent quality control management of the pathological process. Specifically, by combining the pathological image defect quality control detection results and the pathological process abnormality identification results, it automatically identifies the correlation between pathological image quality problems and front-end pathological process abnormalities, monitors and controls the quality of the pathological process in real time, thereby realizing comprehensive quality control and intelligent management of the pathological process.

[0138] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0139] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0140] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A pathological intelligent quality control management system based on big data, characterized in that: It includes a multi-source pathology data collection module, a data optimization module, a pathology image quality control and detection module, a pathology process anomaly tracing module, and an intelligent quality control management module; The multi-source pathological data collection module obtains raw data for pathological quality control management through data acquisition operations; The data optimization module performs pathological image data preprocessing, pathological process management data preprocessing, and multimodal data alignment to obtain optimized pathological quality control management data. The pathological image quality control and detection module is used to automatically detect and identify problems in digital pathological images. Specifically, it first introduces wavelet transform to process multi-source information in the pathological image, combines a dilated convolution aggregation mechanism to achieve global pathological information aggregation, and uses channel attention and spatial attention mechanisms to highlight significant pathological defect areas. A pathological image detection model is then established. The model is trained using historical detection data as training data and a significant pathological region quality control loss function as a supervised target. Finally, real-time detection data is input into the trained model to obtain the pathological image defect quality control and detection results. The pathological process anomaly tracing module is used to trace the abnormal process steps corresponding to pathological image quality problems. Specifically, a pathological process anomaly tracing model is constructed based on a bidirectional long short-term memory neural network, and the model is trained. The search algorithm is improved by introducing a worst individual position iterative correction strategy to obtain the hyperparameter optimization combination of the pathological process anomaly tracing model, and the best-performing pathological process anomaly tracing model is obtained. The real-time management data corresponding to the pathological images and the pathological image defect quality control detection results are input into the model to obtain the pathological process anomaly identification results. The intelligent quality control management module specifically achieves intelligent quality control management of the pathological process by combining the quality control detection results of pathological image defects and the identification results of pathological process anomalies.

2. The pathological intelligent quality control management system based on big data according to claim 1, characterized in that: The pathological image quality control and detection module specifically includes the following steps: Establish a pathological image detection model; The training of the pathological image detection model involves using historical pathological image detection data as training data, employing a pathological salient region quality control loss function as the supervised training objective function, and training the pathological image detection model through gradient descent iterative optimization to finally obtain the trained pathological image detection model. The pathological salient region quality control loss function is used to introduce salient region constraints during the supervised training of the pathological image detection model, making the model pay more attention to salient defect regions in pathological slices during optimization. Specifically, it is obtained by weighting and summing the cross-entropy loss term and the salientity penalty term according to linear dynamic weight parameters; the formula used is as follows: ； ； ； In the formula, This represents the significance penalty weight corresponding to the e-th training epoch. This represents the initial weight value. This represents the maximum weight value. Indicates the current training cycle. Indicates the total number of training cycles. The value represents the significance penalty term, B represents the number of pathological digital images used in the training, and b represents the index of the pathological digital images. This indicates the number of features in this image. The saliency score of the b-th pathological digital image on feature j is calculated using the SHAP method. This represents the quality control loss function value for pathologically significant regions. This represents the value of the cross-entropy loss function; Pathological image defect quality control specifically involves inputting real-time pathological image detection data into a trained pathological image detection model and outputting pathological image defect quality control detection results.

3. The pathological intelligent quality control management system based on big data according to claim 1, characterized in that: The establishment of the pathological image detection model specifically includes the following steps: The multi-source information processing of pathological images involves first performing a two-dimensional discrete wavelet transform on the digital pathological image to decompose it into low-frequency components and high-frequency components in three directions. Then, convolution, normalization, and nonlinear activation are applied to the four types of components to obtain the corresponding sub-band enhancement features. Next, the sub-band enhancement features are reconstructed from the frequency domain to the spatial domain using inverse wavelet transform to obtain the image spatial reconstruction features. Finally, the image spatial reconstruction features are concatenated with the image shallow convolution features in the channel dimension, and the number of channels is compressed through convolution to obtain the final shallow enhancement features. The deep feature extraction of pathological images involves inputting shallow enhancement features into a ResNet50-based neural network, performing initial convolution and layer-by-layer convolution and downsampling in four residual stages to generate a multi-scale deep feature set, and then aligning and compressing the multi-scale feature set through 1×1 convolution to obtain the deep features of the pathological image. The global pathological information aggregation specifically involves inputting deep features of the pathological image into a dilated convolutional aggregation layer, generating the convolutional outputs of each branch, and adding them element-wise with the deep features of the pathological image to obtain global context-enhanced features; the dilated convolutional aggregation layer consists of... It consists of several parallel convolutional branches, each with a different dilation rate; the formula used is as follows: ； In the formula, This represents global context-enhanced features. Indicates deep features in pathological images. This indicates an element-wise addition operation. Indicates the number of convolution branches. Let represent the dilation rate used in the i-th branch, and d represent the dilation coefficient of the convolution. This indicates that the kernel size is 3×3 and the dilation rate is... The dilated convolution operation; The enhancement of pathologically significant defect regions involves taking global context enhancement features as input, first applying a channel attention mechanism to obtain channel attention weighted features, then applying a spatial attention mechanism to the channel attention weighted features to obtain a spatial weight map, and finally weighting the channel attention weighted features in the spatial dimension to obtain significant enhancement features. The defect detection results are output by inputting the fused enhancement features into the convolution classifier head, calculating the probability distribution of each pathological image defect category through the Softmax activation function, and selecting the category corresponding to the highest probability as the pathological image defect detection result.

4. The pathological intelligent quality control management system based on big data according to claim 1, characterized in that: The pathological process anomaly tracing module specifically includes the following steps: A pathological process anomaly tracing model was constructed and trained. Specifically, a pathological process anomaly tracing model was established based on a bidirectional long short-term memory neural network. The historical pathological process management data corresponding to the pathological images and the detection results of historical pathological images were used as training data for the model. The pathological process anomaly tracing model was trained to obtain the trained pathological process anomaly tracing model. Model hyperparameter optimization specifically involves obtaining optimized combinations of hyperparameters for the pathological process anomaly tracing model through an improved search algorithm. The model hyperparameter update specifically involves adjusting the hyperparameters of the pathological process anomaly tracing model based on the optimized combination of the hyperparameters to obtain the optimal-performing pathological process anomaly tracing model. The pathological process anomaly identification specifically involves inputting the real-time pathological process management data corresponding to the pathological image and the pathological image defect quality control detection results into the optimal-performing pathological process anomaly tracing model to obtain the pathological process anomaly identification result.

5. The pathology intelligent quality control management system based on big data according to claim 1, characterized in that: The model hyperparameter optimization specifically includes the following steps: Initialize the search population by encoding the hyperparameters of the trained pathological process anomaly tracing model into search individual position vectors, and generate N search individual position vectors through a random initialization method. Each individual encoding represents a candidate combination of pathological process anomaly tracing model hyperparameters to obtain the initial search population. Population role assignment, specifically, calculating the fitness value of the searched individual in the population. The performance of the pathological process anomaly tracing model based on individual location is used as the fitness value of the individual. Individuals are ranked from best to worst according to their fitness value. The top 20% of search individuals are classified as producers, and the remaining 80% of search individuals are classified as foragers. Producer position updates, specifically based on danger signals. and safety threshold Determine the location and update it accordingly; The worst-performing individual's position is iteratively corrected by employing a periodic oscillation-guided strategy to adjust the position of the worst-performing individual in the population; the formula used is as follows: ； In the formula, This indicates the position of the worst-performing individual in the current iteration after correction. This indicates the position of the worst-performing individual in the current iteration. and Both represent random numbers in the range [0,1]. express Random numbers within a range This indicates the optimal individual position in the current iteration. This indicates the position of the worst-case individual globally. This indicates the maximum number of iterations. The factor representing the control of periodic oscillations is a constant, and its value ranges from [value missing]. ; Forager location updates are specifically based on individual indexes and population size. Danger detectors update their locations by randomly selecting 10% of individuals in the population as danger detectors. The best and worst individuals are updated by recalculating the fitness values ​​of all searched individuals and determining the new global best and worst individual positions based on the fitness values. The search terminates, specifically when the fitness value of the searched individual is... When the fitness level exceeds the fitness threshold or the maximum number of iterations is reached, the search is terminated and the globally optimal individual position is obtained; the globally optimal individual position specifically refers to the optimized combination of hyperparameters.

6. The pathological intelligent quality control management system based on big data according to claim 5, characterized in that: The intelligent quality control management module is used to realize intelligent quality control management of the pathological process. Specifically, by combining the quality control detection results of pathological image defects and the identification results of pathological process anomalies, it automatically identifies the correlation between pathological image quality problems and front-end pathological process anomalies, monitors and controls the quality of the pathological process in real time, thereby realizing comprehensive quality control and intelligent management of the pathological process.

7. The pathological intelligent quality control management system based on big data according to claim 1, characterized in that: The multi-source pathology data collection module specifically obtains raw pathology quality control management data by acquiring data from the pathology laboratory information system and the hospital information system. This raw pathology quality control management data includes historical pathology image detection data, historical pathology process management data, real-time pathology image detection data, and real-time pathology process management data. Both historical and real-time pathology image detection data include digital pathology images. The historical pathology image detection data also includes historical pathology image detection results. Both historical and real-time pathology process management data include pathology equipment operation log data, pathology operation process record data, and pathology operation environment data. The historical pathology process management data also includes operation defect record data.

8. The pathological intelligent quality control management system based on big data according to claim 1, characterized in that: The data optimization module specifically includes the following steps: Preprocessing of pathological image data includes image cleaning, image standardization, image quality enhancement, and image annotation to obtain optimized data for pathological image detection. Preprocessing of pathology workflow management data includes data cleaning, data standardization, and data coding to obtain optimized pathology workflow data. Multimodal data alignment is used to achieve a one-to-one correspondence between the image quality of the same pathological sample and the data of the process steps. Specifically, it is achieved by aligning the pathological image data with the corresponding process management data according to the sample number and timestamp.

Images