Pathological feature recognition and negative elimination method based on microscopic imaging

Abstract

The invention discloses a pathological feature recognition and negative elimination method based on microscopic imaging. The method comprises the following steps: S1, collecting a pathological section image and digitally generating original microscopic image data; s2, preprocessing the original microscopic image; s3, constructing a pathological image recognition network fusing converter coding and a gating dynamic receptive field mechanism, and outputting pathological feature vectors; s4, performing context modeling through an attention guidance and category perception decoder, and outputting an image classification result; s5, constructing a discriminant boundary separation model based on positive and negative sample embedding, and performing negative exclusion judgment; s6, performing confidence coefficient weighted evaluation in combination with the uncertainty and the boundary distance, setting a dynamic threshold value, and screening out low-credibility samples; and S7, coding the classification result and the negative label into structured data, and sending the structured data to a diagnosis auxiliary system. According to the method, multi-scale modeling and a negative screening mechanism are fused, and intelligent recognition and credible diagnosis output of the pathological image are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis and computational pathology, and in particular to a pathological feature recognition and negative exclusion method based on microscopic imaging. Background Art

[0002] With the rapid development of digital pathology, microscopic image analysis has gradually become an important means of assisting pathological diagnosis, playing an important role in tasks such as tumor screening, tissue classification, and morphological feature recognition. Traditional pathological diagnosis relies on manual observation of microscopic sections, which is highly subjective and inefficient. In the face of large-scale image data, it is easily affected by factors such as physician fatigue and experience differences, making it difficult to ensure the accuracy and consistency of diagnostic results. In recent years, with the widespread application of deep learning in the field of medical imaging, more and more studies have attempted to use convolutional neural networks (CNNs) to automatically extract and classify microscopic images. However, existing methods still have obvious limitations when dealing with problems such as multi-scale structures, non-uniform staining, and complex tissue boundaries in pathological images.

[0003] Existing technologies typically use a CNN structure with a fixed receptive field to extract features from pathological images. This makes it difficult to simultaneously capture local morphological features at the cellular level and global spatial relationships at the tissue level, resulting in unstable performance of the model when identifying lesion boundaries or subtle lesions. At the same time, when faced with the problem of fuzzy distribution between pathological image categories or an unbalanced number of samples, existing methods often suffer from insufficient credibility or misclassification of negative samples in classification results. In particular, for the distinction between normal tissue and low-risk lesions, there is often a lack of effective boundary modeling and uncertainty control mechanisms. In addition, existing studies have mostly focused on improving classification accuracy, with less consideration given to how to systematically identify and exclude low-risk or negative samples, thereby optimizing the reading burden of pathologists and improving screening efficiency.

[0004] To address the above issues, the present invention proposes a pathological feature recognition and negative exclusion method based on microscopic imaging. This method uses a recognition network that integrates a transformer encoding module with a gated dynamic receptive field mechanism to achieve collaborative modeling of multi-scale structural features and global contextual information, significantly improving the ability to discriminate under complex tissue structures. At the same time, a discriminant boundary separation model and a confidence-weighted assessment mechanism are constructed to effectively address the shortcomings of existing technologies in negative exclusion and high-uncertainty sample control. Dynamic threshold control enables rapid screening of high-confidence images and review recommendations for low-confidence samples, establishing a complete closed-loop recognition process from image acquisition to structured result output, fundamentally improving the intelligent processing efficiency and diagnostic assistance capabilities of digital pathology images.

[0005] Therefore, how to provide a pathological feature identification and negative exclusion method based on microscopic imaging is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0006] One purpose of the present invention is to propose a pathological feature recognition and negative exclusion method based on microscopic imaging, invent a fusion gated dynamic receptive field and transformer encoding mechanism to achieve multi-scale pathological feature extraction, introduce discriminant boundary separation and confidence weighting strategies, improve the accuracy of negative sample exclusion, construct a structured output and system interface linkage process, and realize intelligent recognition of pathological images, reliable screening and efficient diagnostic assistance.

[0007] A method for pathological feature recognition and negative exclusion based on microscopic imaging according to an embodiment of the present invention includes the following steps:

[0008] S1, obtaining a high-resolution image of a pathological tissue section by a microscopic imaging device, and digitizing the high-resolution image to obtain original microscopic image data;

[0009] S2. performing image preprocessing on the original microscopic image data, including noise suppression, color normalization, tissue boundary enhancement, and foreground region segmentation, to generate a processed image with a clear structure;

[0010] S3. Construct a pathological image recognition network that integrates a transformer encoding module and a gated dynamic receptive field mechanism, takes the processed image as input, extracts multi-scale spatial semantic features through feature alignment, channel dynamic regulation, and global context modeling, and generates a pathological feature vector;

[0011] S4, inputting the pathological feature vector into a classification decoding module, using a multi-layer attention-guided feature fusion structure to perform contextual association modeling on features of different regions, and using a category-aware decoder to generate a classification result of the pathological image;

[0012] S5. Construct a discriminant boundary separation model in the feature compression space, perform negative exclusion judgment on the preliminary classification judgment results based on the positive and negative sample embedding distribution, screen samples whose feature performance is close to the negative template and output them as negative results;

[0013] S6. Perform a confidence-weighted assessment on the negative exclusion results, combine the uncertainty distribution output by the discriminant boundary separation model with the category boundary distance, set a dynamic threshold, filter out low-confidence samples, and retain high-confidence images for subsequent processing;

[0014] S7. Encode the pathological classification results and negative exclusion labels in a unified structural format, write them into a structured data set, and simultaneously send them to the diagnostic assistance system for doctors to call or for remote consultation system analysis.

[0015] Optionally, the S1 specifically includes:

[0016] S11, fixing the stained pathological tissue sections on a glass slide carrier and placing them in a fully automated digital microscope platform to prepare for image acquisition;

[0017] S12, controlling the fully automatic digital microscope platform to automatically scan the tissue slice region by region according to the set scanning magnification and spatial resolution to obtain a local field of view image;

[0018] S13, stitching the multiple local field of view images in order of spatial positions to generate a complete high-resolution slice image;

[0019] S14, performing unified format conversion on the high-resolution slice image to generate original microscopic image data that meets subsequent processing requirements.

[0020] Optionally, the S2 specifically includes:

[0021] S21, performing image denoising processing on the original microscopic image data, using edge-preserving filtering to remove background noise and imaging artifacts, and retaining tissue structure details;

[0022] S22, performing color normalization processing on the staining conditions in the original microscopic image, unifying the color distribution of the image based on the target staining template, and improving the consistency between images;

[0023] S23, performing tissue structure edge enhancement operation on the boundary area of ​​the original microscopic image, strengthening the cell and tissue contours through gradient operation, and highlighting the local structural features;

[0024] S24. Use a foreground extraction algorithm to segment the original microscopic image, remove the background area and extract the tissue foreground area, and generate a processed image with clear structure and clear boundaries.

[0025] Optionally, the S3 specifically includes:

[0026] S31, inputting the processed image obtained in step S2 into a shallow feature extraction module, wherein the shallow feature extraction module is composed of multiple standard two-dimensional convolutional layers, batch normalization layers and nonlinear activation functions, and is used to extract basic visual features such as low-level edges, textures and color distributions of the image to generate a primary feature map;

[0027] S32. Input the primary feature map into a gated dynamic receptive field module. The gated dynamic receptive field module is composed of multiple parallel convolution branches, each branch has a different convolution kernel size and stride parameter, and is used to extract multi-scale structural features. The gating unit adaptively assigns weights to each branch according to the channel response value, and fuses and outputs a multi-scale feature map.

[0028] S33, inputting the multi-scale feature map into a feature enhancement module, wherein the feature enhancement module includes a channel attention mechanism and a spatial attention mechanism, weightedly enhancing high-response channels, strengthening the spatial expression ability of the lesion area, and outputting an enhanced feature map;

[0029] S34, inputting the enhanced feature map into a transformer encoding module, wherein the transformer encoding module is composed of a multi-layer stacked encoder unit, each encoder unit including a multi-head self-attention calculation layer, a feedforward network and a residual connection structure, while introducing relative position encoding to maintain the spatial semantics of the pathological structure, extracting context information between distant structures, and outputting a context encoding feature map;

[0030] S35. Construct a cross-scale interactive attention module to align the scales of the encoded feature maps of different depths and establish a cross-attention map. Calculate the weighted similarity between scales through the feature matching function, fuse the contextual connections between high-expression areas and low-expression areas, and output a cross-scale fused feature map.

[0031] S36. Input the cross-scale fusion feature map into the structure-semantic separation unit to separate the main path features for structure preservation and the auxiliary path features for semantic enhancement. The main path uses the position attention mechanism to enhance structural stability, and the auxiliary path uses the context enhancement mechanism to strengthen the semantic representation of the lesion area. The paths are then weighted and combined through the path fusion module.

[0032] S37, performing a dimension compression mapping operation on the fused feature map, converting the multi-dimensional image features into a vector representation of uniform dimension through global average pooling and linear transformation, forming a final pathological image feature vector and outputting it;

[0033] Optionally, the S4 specifically includes:

[0034] S41, receiving the pathological image feature vector output in step S3, and expanding the feature vector into a two-dimensional spatial feature map through a linear projection layer for subsequent structural reconstruction and classification reasoning;

[0035] S42, inputting the two-dimensional spatial feature map into a multi-level classification decoding network, wherein the multi-level classification decoding network is composed of a plurality of stacked attention fusion decoding units, each of which includes an attention fusion module, a convolution restoration module, and a channel fusion module;

[0036] S43. In each attention fusion module, a joint modeling mechanism of channel attention and spatial attention is introduced. Highly correlated feature channels are dynamically enhanced through global pooling and local response adjustment strategies. The spatial position information is used to calculate the salient area attention map to achieve spatially guided feature focusing.

[0037] S44, input the attention-enhanced feature map into a context-aware fusion module, the context-aware fusion module includes a local context aggregation sub-module and a non-local relationship capture sub-module, the local context aggregation sub-module calculates the neighborhood semantic consistency based on a sliding window strategy, and the non-local relationship capture sub-module calculates the global semantic correlation between distant regions based on pixel-level feature similarity;

[0038] S45, input the context-aware fused feature map into a category-aware decoder, the category-aware decoder includes a category embedding layer, a feature matching layer and a category response generation layer, the category embedding layer maps the preset pathological tissue type into a high-dimensional embedding vector, the feature matching layer calculates the correlation score between each pixel or region feature and the category embedding, and the category response generation layer outputs a preliminary classification probability map according to the correlation score;

[0039] S46, on the basis of the output of the category-aware decoder, a residual enhancement channel is introduced to guide the fusion of deep semantic features and shallow structural features, enhance the classification boundary discrimination, and retain the tissue morphology details to prevent feature information from being excessively smoothed;

[0040] S47, scale consistency adjustment is performed on the preliminary classification probability map, a multi-resolution fusion strategy is introduced, the output features of different spatial levels are up-sampled, normalized and weighted fused to form a multi-channel classification map of uniform scale;

[0041] S48, input the multi-channel classification map into an integrated decision module, the module includes a plurality of parallel classification heads, each classification head performs independent discrimination based on different feature branches, and generates a final classification decision vector by using weight weighting;

[0042] S49, the final classification decision vector is subjected to Softmax normalization processing, and the final classification result of the pathological image is output, which is used as the input basic data of the subsequent negative exclusion module.

[0043] Optionally, the S5 specifically includes:

[0044] S51, receive the pathological image classification decision result and the corresponding pathological image feature vector generated in step S4, and input the pathological image feature vector into a feature compression mapping module, the feature compression mapping module includes a full connection layer, a normalization layer and a dimension reduction transformation layer, and is used for mapping the original high-dimensional feature vector to a low-dimensional discriminant space;

[0045] S52, construct a positive sample embedding distribution and a negative sample embedding distribution in the low-dimensional discriminant space, the positive sample embedding distribution is clustered according to the features of known positive cases, and the negative sample embedding distribution is constructed according to the boundary template of the features of known negative cases;

[0046] S53. Based on the embedding distribution of the positive samples and the negative samples, construct a discriminant boundary separation model, wherein the discriminant boundary separation model includes an inter-class spacing constraint mechanism and a boundary direction enhancement mechanism, which is used to form a distinguishable feature region boundary in a low-dimensional space;

[0047] S54, performing distance matching on the feature points of the image to be judged and the negative sample template in the discrimination space, using a distance weight function to evaluate the closeness to the negative distribution, and generating a discrimination similarity score;

[0048] S55, inputting the discriminant similarity score into the negative probability scoring module, and generating a multi-factor evaluation vector by combining the category boundary separation strength and the discriminant confidence information;

[0049] S56. Input the multi-factor evaluation vector into a negative exclusion determination module, which uses a decision tree strategy to logically combine the determination conditions, identify samples that meet the negative distribution characteristics, and output them as negative results;

[0050] S57: Mark the image samples that are determined to be negative as "excluded" and send them to the downstream module for archiving or recording, and do not enter the subsequent in-depth analysis process;

[0051] S58. Image samples judged as non-negative are retained for subsequent confidence assessment or manual review to ensure that the complete diagnostic process of high-risk or uncertain samples can be executed.

[0052] Optionally, the S6 specifically includes:

[0053] S61, receiving the negative exclusion result output in step S5 and its corresponding discriminant similarity score and multi-factor evaluation vector, and constructing a confidence vector representation for each sample;

[0054] S62: Input the confidence vector into a confidence weighted evaluation module, which is composed of a boundary proximity evaluation unit, an intra-class consistency scoring unit, and a model uncertainty measurement unit, and respectively calculates the shortest boundary distance between the sample and the negative template, its local distribution consistency score within the class, and its prediction output stability index;

[0055] S63. Perform weighted fusion on the three confidence score results and generate a single comprehensive confidence score using a feature weighted aggregation mechanism. The range of the single comprehensive confidence score value is normalized to a fixed interval as the final representation of the sample credibility.

[0056] S64. Construct a confidence-boundary coupling mapping model, perform bivariate modeling on the relationship between the comprehensive confidence score and the boundary position of the sample in the discriminant space, and obtain the uncertainty distribution state of the sample;

[0057] S65. Dynamically adjust the threshold setting strategy based on the uncertainty distribution state and the sample category label, wherein the threshold setting strategy is based on the category density estimate and the confidence fluctuation range to determine the minimum credible judgment standard under different categories;

[0058] S66: Compare the dynamic threshold with the comprehensive confidence score, mark samples below the threshold as low-confidence samples, and output them to the manual review processing path;

[0059] S67: Mark samples with a value higher than the dynamic threshold as high-confidence samples, and directly enter the subsequent processing step or output to the diagnosis report module;

[0060] S68. Record all parameters and scoring results involved in threshold determination, write them into the evaluation log file, and use them as a data source for continuous learning and parameter adjustment of the system model;

[0061] S69. Perform confidence hierarchical classification on all retained samples and establish three output levels of high confidence, medium confidence and low confidence for subsequent path control and processing priority allocation strategy.

[0062] Optionally, the S7 specifically includes:

[0063] S71, receiving the pathological image classification determination results and negative exclusion results outputted in step S4 and step S6, respectively, and generating a comprehensive information vector including a category label, determination confidence, negative status mark, and evaluation log path for each image sample;

[0064] S72. Input the comprehensive information vector into a result encoding module. The result encoding module includes a field mapping unit, a format standardization unit, and a structure generation unit. The field mapping unit extracts required information based on predefined field templates, the format standardization unit standardizes various data formats to target encoding rules, and the structure generation unit creates standard structured record rows based on primary keys.

[0065] S73. Writing the structured results into a pathology diagnosis database. The pathology diagnosis database adopts a multi-table joint structure. The main table records the image unique identifier and the final classification result, and the appendix records the evaluation parameters, negative judgment details, and historical recognition process metadata.

[0066] S74. Synchronously sending the written data to the diagnosis assistance system interface. The diagnosis assistance system interface uses a standard data transmission protocol and supports data interaction with the hospital information system, imaging workstation, manual review terminal, or remote collaboration platform.

[0067] S75. Introduce encryption mechanisms and identity verification processes during data transmission to ensure that image recognition results are not tampered with or leaked during transmission;

[0068] S76. Generate an automatic recognition record summary for each classification result and negative exclusion mark, and push it to the doctor's work interface for pre-judgment suggestions and quick review reference;

[0069] S77. Set up a structured output interface log function to record and archive the time, target address, content summary, and feedback status of each result output for subsequent performance evaluation and system retrospective analysis.

[0070] The beneficial effects of the present invention are:

[0071] By integrating microscopic imaging and deep semantic modeling technology, this paper proposes a feature recognition and negative exclusion method for pathological images, which significantly improves the automation level and diagnostic efficiency of pathological image analysis.

[0072] First, during the image feature extraction phase, the present invention introduces a recognition network that integrates a transformer encoding module with a gated dynamic receptive field mechanism. This overcomes the issues of traditional convolutional architectures, such as a fixed receptive field and insufficient context modeling. This network adaptively adjusts the feature extraction scale based on image content, effectively capturing organizational features at different spatial levels and improving the recognition accuracy of the model in complex structural areas.

[0073] Secondly, the present invention establishes a negative exclusion strategy that combines a category-aware decoder with a discriminant boundary separation model, enabling a more refined distinction between sample categories based on feature space distribution. By introducing a low-dimensional discriminant space and a positive-negative sample embedding mechanism, it achieves accurate identification and automatic exclusion of negative samples, effectively reducing the manual screening workload for doctors faced with large numbers of normal images and reducing diagnostic redundancy.

[0074] Furthermore, the present invention establishes a confidence-weighted credibility assessment mechanism that comprehensively considers multiple factors, including boundary distance, intra-class consistency, and model uncertainty, to dynamically set the classification credibility threshold, significantly improving the ability to identify low-confidence samples. This mechanism not only enhances the system's responsiveness to abnormal or marginal samples but also provides a priority basis for subsequent manual review and intelligent diagnosis, optimizing the allocation of diagnostic resources.

[0075] Finally, this invention uniformly encodes classification results and negative markers into structured data, seamlessly integrating them with hospital diagnostic support systems or remote consultation platforms. This truly automates the entire process from image acquisition, processing, and recognition to structured output, demonstrating excellent scalability and clinical application value. Overall, this invention significantly outperforms existing technologies in terms of image recognition accuracy, negative exclusion capability, system robustness, and structured output efficiency, providing an effective technical path for the intelligent development of digital pathology. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0077] Figure 1 This is an overall flow chart of a method for pathological feature recognition and negative exclusion based on microscopic imaging proposed by the present invention;

[0078] Figure 2 Schematic diagram of the pathological image recognition network structure that integrates transformer encoding and gated dynamic receptive field mechanism proposed in the present invention;

[0079] Figure 3 This is a flow chart of the dynamic threshold screening of the confidence weighted evaluation mechanism proposed in the present invention. DETAILED DESCRIPTION

[0080] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0081] refer to Figure 1-3 , a method for pathological feature identification and negative exclusion based on microscopic imaging, comprising the following steps:

[0082] S1, obtaining a high-resolution image of a pathological tissue section by a microscopic imaging device, and digitizing the high-resolution image to obtain original microscopic image data;

[0083] S2. performing image preprocessing on the raw microscopic image data, including noise suppression, color normalization, tissue boundary enhancement, and foreground region segmentation, to generate a processed image with a clear structure;

[0084] S3. Construct a pathological image recognition network that integrates a transformer encoding module and a gated dynamic receptive field mechanism, takes the processed image as input, extracts multi-scale spatial semantic features through feature alignment, channel dynamic regulation, and global context modeling, and generates a pathological feature vector;

[0085] S4, inputting the pathological feature vector into a classification decoding module, using a multi-layer attention-guided feature fusion structure to perform contextual association modeling on features of different regions, and using a category-aware decoder to generate a classification result of the pathological image;

[0086] S5. Construct a discriminant boundary separation model in the feature compression space, perform negative exclusion judgment on the preliminary classification judgment results based on the positive and negative sample embedding distribution, screen samples whose feature performance is close to the negative template and output them as negative results;

[0087] S6. Perform a confidence-weighted assessment on the negative exclusion results, combine the uncertainty distribution output by the discriminant boundary separation model with the category boundary distance, set a dynamic threshold, filter out low-confidence samples, and retain high-confidence images for subsequent processing;

[0088] S7. Encode the pathological classification results and negative exclusion labels in a unified structural format, write them into a structured data set, and simultaneously send them to the diagnostic assistance system for doctors to call or for remote consultation system analysis.

[0089] The present invention constructs an integrated method for intelligent recognition and negative exclusion of pathological images, covering the entire process from image acquisition, preprocessing, multi-scale feature extraction, attention fusion classification, feature boundary judgment to credible screening and structured output. Compared with the existing single classification model, this method structurally introduces deep transformer encoding and dynamic receptive field mechanism collaborative modeling, which improves the ability to characterize lesion areas and complex tissue morphology. At the same time, by discriminating boundary separation and confidence assessment mechanisms, it effectively solves the problems of inaccurate exclusion of negative samples and high misdiagnosis rate in traditional methods, realizes the automatic closed loop of high-precision recognition and screening processes, greatly reduces the pressure of manual review, and improves the intelligence level and stability of the auxiliary diagnosis system.

[0090] In this embodiment, S1 specifically includes:

[0091] S11, fixing the stained pathological tissue sections on a glass slide carrier and placing them in a fully automated digital microscope platform to prepare for image acquisition;

[0092] S12, controlling the fully automatic digital microscope platform to automatically scan the tissue slice region by region according to the set scanning magnification and spatial resolution to obtain a local field of view image;

[0093] S13, stitching the multiple local field of view images in order of spatial positions to generate a complete high-resolution slice image;

[0094] S14, performing unified format conversion on the high-resolution slice image to generate original microscopic image data that meets subsequent processing requirements.

[0095] By introducing a fully automated digital microscope platform, high-throughput, standardized slice image acquisition and stitching are achieved, and image formats are standardized, providing highly consistent, high-quality input data for subsequent model training and deployment. Compared to traditional manual image acquisition methods, this method eliminates issues such as inconsistent image resolution and scan overlap errors caused by human error, ensuring the integrity and accuracy of the original image information, effectively supporting the model's robust extraction of minute structural features, and improving overall system stability.

[0096] In this embodiment, S2 specifically includes:

[0097] S21, performing image denoising processing on the original microscopic image data, using edge-preserving filtering to remove background noise and imaging artifacts, and retaining tissue structure details;

[0098] S22, performing color normalization processing on the staining conditions in the original microscopic image, unifying the color distribution of the image based on the target staining template, and improving the consistency between images;

[0099] S23, performing tissue structure edge enhancement operation on the boundary area of ​​the original microscopic image, strengthening the cell and tissue contours through gradient operation, and highlighting the local structural features;

[0100] S24. Use a foreground extraction algorithm to segment the original microscopic image, remove the background area and extract the tissue foreground area, and generate a processed image with clear structure and clear boundaries.

[0101] This method integrates multiple enhancement and segmentation strategies in the image preprocessing phase, from edge-preserving filtering, color normalization, structural boundary enhancement, to foreground extraction, to construct a processing pipeline tailored to the characteristics of stained pathology images. This processing flow can significantly reduce interference from non-tissue regions, improve the separation of cellular structures within the image, and provide clearer input features for subsequent recognition networks. Compared with traditional processing methods that only use color correction or edge extraction, this method can effectively adapt to image variability under multi-batch and multi-device acquisition conditions, significantly enhancing the model's ability to transfer across cross-domain tasks.

[0102] In this embodiment, S3 specifically includes:

[0103] S31, inputting the processed image obtained in step S2 into a shallow feature extraction module, wherein the shallow feature extraction module is composed of multiple standard two-dimensional convolutional layers, batch normalization layers and nonlinear activation functions, and is used to extract basic visual features such as low-level edges, textures and color distributions of the image to generate a primary feature map;

[0104] S32. Input the primary feature map into a gated dynamic receptive field module. The gated dynamic receptive field module is composed of multiple parallel convolution branches, each branch has a different convolution kernel size and stride parameter, and is used to extract multi-scale structural features. The gating unit adaptively assigns weights to each branch according to the channel response value, and fuses and outputs a multi-scale feature map.

[0105] S33, inputting the multi-scale feature map into a feature enhancement module, wherein the feature enhancement module includes a channel attention mechanism and a spatial attention mechanism, weightedly enhancing high-response channels, strengthening the spatial expression ability of the lesion area, and outputting an enhanced feature map;

[0106] S34, input the enhanced feature map into a transformer encoder module, the transformer encoder module is composed of a plurality of stacked encoder units, each encoder unit includes a multi-head self-attention calculation layer, a feedforward network and a residual connection structure, while introducing relative position encoding to maintain the spatial semantics of pathological structures, extract the context information between long-distance structures, and output a context encoding feature map;

[0107] S35, construct a cross-scale interaction attention module, scale align the encoding feature maps of different depths and establish cross attention mapping, calculate the similarity weighting value between scales through a feature matching function, fuse the context relationship between high expression regions and low expression regions, and output a cross-scale fusion feature map;

[0108] S36, input the cross-scale fusion feature map into a structure semantic separation unit, separate the main path feature for structure preservation and the auxiliary path feature for semantic enhancement, the main path uses a position attention mechanism to enhance the structure stability, the auxiliary path uses a context enhancement mechanism to enhance the semantic representation of the lesion area, and then the path fusion module is used for weighted combination;

[0109] S37, perform dimension compression mapping operation on the fused feature map, convert the multi-dimensional image feature into a unified dimensional vector representation through global average pooling and linear transformation, construct a final pathological image feature vector and output;

[0110] The recognition network architecture fuses the transformer encoder module and the gated dynamic receptive field mechanism, breaking the limitations of fixed receptive field, insufficient long-distance dependence expression of traditional convolutional neural networks. In the model, the gated branch structure guides multi-scale path selection, and adapts to different structure scale information; the transformer module efficiently models the long-distance semantic relationship between organizations through global attention mechanism, thereby greatly improving the lesion recognition accuracy. Especially in dealing with pathological images with strong morphological heterogeneity and fuzzy region boundaries, this network structure can better capture high-order features and global context semantics, greatly improving the model's ability to distinguish tissue types and boundaries.

[0111] In the embodiment, the S4 specifically comprises:

[0112] S41, receive the pathological image feature vector output in step S3, and expand the feature vector into a two-dimensional space feature map through a linear projection layer, for subsequent structure reconstruction and classification reasoning;

[0113] S42, input the two-dimensional space feature map into a multi-level classification decoding network, the multi-level classification decoding network is composed of a plurality of stacked attention fusion decoding units, each attention fusion decoding unit includes an attention fusion module, a convolution restoration module and a channel fusion module;

[0114] S43. In each attention fusion module, a joint modeling mechanism of channel attention and spatial attention is introduced. Highly correlated feature channels are dynamically enhanced through global pooling and local response adjustment strategies. The spatial position information is used to calculate the salient area attention map to achieve spatially guided feature focusing.

[0115] S44, inputting the attention-enhanced feature map into a context-aware fusion module, wherein the context-aware fusion module includes a local context aggregation submodule and a non-local relationship capture submodule, wherein the local context aggregation submodule calculates neighborhood semantic consistency based on a sliding window strategy, and the non-local relationship capture submodule calculates global semantic correlation between distant regions based on pixel-level feature similarity;

[0116] S45. Inputting the context-aware fused feature map into a category-aware decoder, wherein the category-aware decoder includes a category embedding layer, a feature matching layer, and a category response generation layer. The category embedding layer maps the preset pathological tissue type into a high-dimensional embedding vector. The feature matching layer calculates the correlation score between each pixel or region feature and the category embedding. The category response generation layer outputs a preliminary classification probability map based on the correlation score.

[0117] S46. Based on the output of the category-aware decoder, a residual enhancement path is introduced to guide the fusion of deep semantic features and shallow structural features, enhance the classification boundary discrimination, while preserving tissue morphological details and preventing over-smoothing of feature information;

[0118] S47. Adjust the scale consistency of the preliminary classification probability map, introduce a multi-resolution fusion strategy, upsample, normalize and weightedly fuse the output features of different spatial levels to form a multi-channel classification map of unified scale;

[0119] S48, inputting the multi-channel classification map into an integrated decision module, which includes multiple parallel classification heads, each of which performs independent discrimination based on different feature branches and generates a final classification decision vector using a weighted method;

[0120] S49: The final classification decision vector is subjected to Softmax normalization processing, and the final classification result of the pathological image is output as the input basic data of the subsequent negative exclusion module.

[0121] By constructing a discriminant boundary separation model in the feature compression space, the embedding of pathological image samples in the low-dimensional feature space and classification boundary modeling are achieved, effectively distinguishing between positive and negative samples. In traditional classification methods based on probability judgment, negative samples are easily misidentified as positive, especially when the feature distribution boundary is fuzzy. This method introduces a boundary direction enhancement mechanism and inter-class distance constraints to construct a clear negative template. The discriminant similarity is calculated through feature matching, which greatly improves the accuracy of negative sample recognition, reduces the risk of misdiagnosis, and brings safer application value to the pathology screening system.

[0122] In this embodiment, the S5 specifically includes:

[0123] S51, receiving the pathological image classification determination result generated in step S4 and its corresponding pathological image feature vector, and inputting the pathological image feature vector into a feature compression mapping module, wherein the feature compression mapping module includes a fully connected layer, a normalization layer, and a dimensionality reduction transformation layer, and is used to map the original high-dimensional feature vector into a low-dimensional discriminant space;

[0124] S52. Constructing a positive sample embedding distribution and a negative sample embedding distribution in a low-dimensional discriminant space, wherein the positive sample embedding distribution is clustered according to the characteristics of known positive cases, and the negative sample embedding distribution is constructed as a boundary template according to the characteristics of known negative cases;

[0125] S53. Based on the embedding distribution of the positive samples and the negative samples, construct a discriminant boundary separation model, wherein the discriminant boundary separation model includes an inter-class spacing constraint mechanism and a boundary direction enhancement mechanism, which is used to form a distinguishable feature region boundary in a low-dimensional space;

[0126] S54, performing distance matching on the feature points of the image to be judged and the negative sample template in the discrimination space, using a distance weight function to evaluate the closeness to the negative distribution, and generating a discrimination similarity score;

[0127] S55, inputting the discriminant similarity score into the negative probability scoring module, and generating a multi-factor evaluation vector by combining the category boundary separation strength and the discriminant confidence information;

[0128] S56. Input the multi-factor evaluation vector into a negative exclusion determination module, which uses a decision tree strategy to logically combine the determination conditions, identify samples that meet the negative distribution characteristics, and output them as negative results;

[0129] S57: Mark the image samples that are determined to be negative as "excluded" and send them to the downstream module for archiving or recording, and do not enter the subsequent in-depth analysis process;

[0130] S58. Image samples judged as non-negative are retained for subsequent confidence assessment or manual review to ensure that the complete diagnostic process of high-risk or uncertain samples can be executed.

[0131] The present invention integrates three-dimensional information of boundary proximity, intra-class consistency score, and model prediction uncertainty through a confidence-weighted evaluation mechanism to comprehensively measure the credibility of sample classification results. Compared with the traditional simplified judgment mechanism based on fixed thresholds or Softmax probability values, this method dynamically adjusts the threshold strategy and constructs a coupled mapping model to achieve refined credibility control for different sample categories and distribution characteristics. This mechanism not only improves the system's response sensitivity to uncertain samples, but also ensures the interpretability and risk controllability of recognition results. It is particularly suitable for clinical scenarios with extremely low tolerance for misjudgment.

[0132] In this embodiment, S6 specifically includes:

[0133] S61, receiving the negative exclusion result output in step S5 and its corresponding discriminant similarity score and multi-factor evaluation vector, and constructing a confidence vector representation for each sample;

[0134] S62: Input the confidence vector into a confidence weighted evaluation module, which is composed of a boundary proximity evaluation unit, an intra-class consistency scoring unit, and a model uncertainty measurement unit, and respectively calculates the shortest boundary distance between the sample and the negative template, its local distribution consistency score within the class, and its prediction output stability index;

[0135] S63. Perform weighted fusion on the three confidence score results and generate a single comprehensive confidence score using a feature weighted aggregation mechanism. The range of the single comprehensive confidence score value is normalized to a fixed interval as the final representation of the sample credibility.

[0136] S64. Construct a confidence-boundary coupling mapping model, perform bivariate modeling on the relationship between the comprehensive confidence score and the boundary position of the sample in the discriminant space, and obtain the uncertainty distribution state of the sample;

[0137] S65. Dynamically adjust the threshold setting strategy based on the uncertainty distribution state and the sample category label, wherein the threshold setting strategy is based on the category density estimate and the confidence fluctuation range to determine the minimum credible judgment standard under different categories;

[0138] S66: Compare the dynamic threshold with the comprehensive confidence score, mark samples below the threshold as low-confidence samples, and output them to the manual review processing path;

[0139] S67: Mark samples with a value higher than the dynamic threshold as high-confidence samples, and directly enter the subsequent processing step or output to the diagnosis report module;

[0140] S68. Record all parameters and scoring results involved in threshold determination, write them into the evaluation log file, and use them as a data source for continuous learning and parameter adjustment of the system model;

[0141] S69. Perform confidence hierarchical classification on all retained samples and establish three output levels of high confidence, medium confidence and low confidence for subsequent path control and processing priority allocation strategy.

[0142] This invention structures the pathology image classification results and negative exclusion markers, and enables efficient integration with diagnostic assistance systems, completing the standardized output and automated integration of diagnostic data. By encapsulating multi-field results and standardizing transmission protocols, this method effectively addresses the issues of inconsistent data interfaces and inconvenient result access in traditional pathology systems. It provides a solid data foundation for remote consultations, model decision review, and system integration, significantly improving the deployability and information interoperability of intelligent recognition systems in actual clinical workflows.

[0143] In this embodiment, the S7 specifically includes:

[0144] S71, receiving the pathological image classification determination results and negative exclusion results outputted in step S4 and step S6, respectively, and generating a comprehensive information vector including a category label, determination confidence, negative status mark, and evaluation log path for each image sample;

[0145] S72. Input the comprehensive information vector into a result encoding module. The result encoding module includes a field mapping unit, a format standardization unit, and a structure generation unit. The field mapping unit extracts required information based on predefined field templates, the format standardization unit standardizes various data formats to target encoding rules, and the structure generation unit creates standard structured record rows based on primary keys.

[0146] S73. Writing the structured results into a pathology diagnosis database. The pathology diagnosis database adopts a multi-table joint structure. The main table records the image unique identifier and the final classification result, and the appendix records the evaluation parameters, negative judgment details, and historical recognition process metadata.

[0147] S74. Synchronously sending the written data to the diagnosis assistance system interface. The diagnosis assistance system interface uses a standard data transmission protocol and supports data interaction with the hospital information system, imaging workstation, manual review terminal, or remote collaboration platform.

[0148] S75. Introduce encryption mechanisms and identity verification processes during data transmission to ensure that image recognition results are not tampered with or leaked during transmission;

[0149] S76. Generate an automatic recognition record summary for each classification result and negative exclusion mark, and push it to the doctor's work interface for pre-judgment suggestions and quick review reference;

[0150] S77. Set up a structured output interface log function to record and archive the time, target address, content summary, and feedback status of each result output for subsequent performance evaluation and system retrospective analysis.

[0151] By building a complete closed-loop mechanism for image acquisition, preprocessing, feature extraction, classification and judgment, negative screening, and credibility assessment, this method integrates image processing and diagnostic tasks, supporting an intelligent and automated flow from front-end acquisition to back-end result output. Compared to the traditional process, which involves multiple separate modules and a high reliance on manual judgment, this method achieves a smooth end-to-end data link and optimized structure, improving diagnostic efficiency while ensuring the traceability and controllability of key judgment links. This provides a standardized, automated, and highly accurate technical path for building an intelligent pathology platform.

[0152] Example 1:

[0153] In order to verify the feasibility of the present invention in implementation, the present invention was applied to the daily work of the pathology department of a certain tertiary hospital. Doctors need to process a large number of microscopic images of pathological samples such as breast biopsy, lung tissue puncture, and lymph node sections every day. After staining, these images show a complex and variable tissue structure. Especially in the early screening stage, a large number of images are negative or suspected to be negative. Relying solely on manual review one by one is not only a huge workload, but also easily interfered by subjective judgment, fatigue level and other factors, resulting in misjudgment or missed judgment. In order to relieve the pressure on pathologists and improve the efficiency of negative screening and the accuracy of diagnosis, the pathology department introduced a pathological feature recognition and negative exclusion method based on microscopic imaging proposed by the present invention, and systematically upgraded its digital pathology workflow.

[0154] In the application scenario, the doctor first places the stained pathological tissue section on the fully automatic digital microscope platform, and automatically scans the entire slide area by area by setting the scanning magnification and resolution parameters. After the scan is completed, the system seamlessly stitches hundreds of local images to generate a complete high-resolution slice image, and uniformly converts it into a 25600×15360 pixel TIFF format. Subsequently, these images are passed to the image preprocessing module. At this stage, the system automatically performs noise removal, color normalization, edge enhancement, and tissue area segmentation operations. Compared with the unprocessed image, the preprocessed image has an improvement of 22.5% in edge clarity and a reduction of nearly 30% in background artifacts, making it more suitable for subsequent feature extraction and analysis.

[0155] After image processing is complete, the system activates a recognition network that integrates a transformer encoding module with a gated dynamic receptive field mechanism. After extracting basic features through shallow convolution, the gated mechanism dynamically selects different receptive field paths to extract multi-scale structural information. Combined with the transformer module, it then models long-range semantic relationships between different tissue regions. Experimental data shows that this network architecture improves lesion recognition accuracy by approximately 8.7% compared to the traditional ResNet architecture. In experiments with breast biopsy images, the model achieved an accuracy of 94.3% for ductal carcinoma regions and 91.6% for atypical hyperplasia regions.

[0156] During the classification phase, the model further inputs features into a multi-layer attention-guided decoder, which outputs preliminary classification results through context modeling and category-aware mechanisms. The system constructs a negative template embedding distribution and boundary separation model, using a low-dimensional feature space to exclude negative samples. In nearly a month of use, the system processed 21,856 breast pathology slides, of which 62.4% were negative. By discriminating boundary screening, the system successfully excluded 12,294 high-confidence negative images, reducing the manual reading burden by more than half.

[0157] To avoid misclassifying low-confidence negative images, the system incorporates a confidence weighting mechanism. This mechanism performs a weighted fusion of boundary distance, uncertainty score, and intra-class consistency assessment on all excluded samples, and outputs a confidence level based on a dynamic threshold strategy. This mechanism has been proven to improve the negative fidelity rate to 96.1% for negative identification samples, effectively preventing false negatives from reaching the subsequent reporting stage. For low-confidence samples, the system automatically marks them as "requiring manual review," ensuring that the final diagnosis of high-risk images is made by experienced physicians.

[0158] The system ultimately writes the classification results, negative labels, and confidence levels into the hospital's diagnostic database as structured data, automatically synchronizing them to the remote consultation platform and auxiliary diagnosis terminals. Doctors can visually view the model's predicted results for each slide, along with regional annotations, confidence scores, and judgment basis, on the electronic pathology platform, achieving an efficient collaborative mechanism of "preemptive screening + critical review."

[0159] Overall, the introduction of this system has reduced the pathology department's workload during the initial image screening phase by 41.7%, shortening the overall workflow by an average of approximately 2.6 hours per day. This has significantly alleviated diagnostic bottlenecks during holidays and concentrated screening periods. Furthermore, physicians consistently report satisfaction with the system's recognition accuracy exceeding 92%, with the system generally believed to help reduce repetitive operations, focus on high-risk areas, and improve the efficiency and controllability of work processes.

[0160] In summary, this embodiment clearly demonstrates the outstanding value of the present invention in improving the efficiency of negative screening, reducing the misjudgment rate, and supporting structured diagnostic flow. It effectively solves the technical problems of low efficiency, large fluctuations in accuracy, and concentrated review pressure in the traditional pathological analysis process, and has significant clinical application prospects.

[0161] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for pathological feature identification and negative exclusion based on microscopic imaging, characterized in that: The steps include: S1, obtaining a high-resolution image of a pathological tissue section by a microscopic imaging device, and digitizing the high-resolution image to obtain original microscopic image data; S2. performing image preprocessing on the original microscopic image data to generate a processed image with a clear structure; S3. Construct a pathological image recognition network that integrates a transformer encoding module and a gated dynamic receptive field mechanism, takes the processed image as input, extracts multi-scale spatial semantic features through feature alignment, channel dynamic regulation, and global context modeling, and generates a pathological feature vector; S4. Inputting the pathological feature vector into a classification decoding module, using a multi-layer attention-guided feature fusion structure to perform contextual association modeling on features of different regions, and using a category-aware decoder to generate a preliminary classification result of the pathological image; S5. Construct a discriminant boundary separation model in the feature compression space, perform negative exclusion judgment on the preliminary classification judgment results based on the positive and negative sample embedding distribution, screen samples whose feature performance is close to the negative template and output them as negative results; S6. Perform a confidence-weighted assessment on the negative exclusion results, combining the uncertainty distribution output by the discriminant boundary separation model with the class boundary distance, and set a dynamic threshold; S7. Encode the pathological classification results and negative exclusion labels in a unified structural format, write them into a structured data set, and send them to the diagnostic assistance system simultaneously.

2. The method for pathological feature identification and negative exclusion based on microscopic imaging according to claim 1, characterized in that: Said S1 specifically includes: S11, fixing the stained pathological tissue sections on a glass slide carrier and placing them in a fully automated digital microscope platform to prepare for image acquisition; S12, controlling the fully automatic digital microscope platform to automatically scan the tissue slice region by region according to the set scanning magnification and spatial resolution to obtain a local field of view image; S13, stitching the multiple local field of view images in order of spatial positions to generate a complete high-resolution slice image; S14. Convert the high-resolution slice image into a unified format to generate original microscopic image data.

3. The method for pathological feature identification and negative exclusion based on microscopic imaging according to claim 1, characterized in that: The S2 specifically includes: S21, performing image denoising processing on the original microscopic image data, using edge-preserving filtering to remove background noise and imaging artifacts, and retaining tissue structure details; S22, performing color normalization processing on the staining condition in the original microscopic image, and unifying the color distribution of the image based on the target staining template; S23, performing tissue structure edge enhancement operation on the boundary area of ​​the original microscopic image, strengthening the cell and tissue contours through gradient operation, and highlighting the local structural features; S24. Use a foreground extraction algorithm to segment the original microscopic image, remove the background area and extract the tissue foreground area, and generate a processed image with clear structure and clear boundaries.

4. The method for pathological feature identification and negative exclusion based on microscopic imaging according to claim 1, characterized in that: The S3 specifically includes: S31, inputting the processed image obtained in step S2 into a shallow feature extraction module, wherein the shallow feature extraction module is composed of multiple standard two-dimensional convolutional layers, multiple normalization layers and non-linear activation functions, and extracting basic visual features including low-level edges, textures and color distributions of the image to generate a primary feature map; S32. Input the primary feature map into a gated dynamic receptive field module. The gated dynamic receptive field module is composed of multiple parallel convolution branches, each branch has a different convolution kernel size and stride parameter, extracts multi-scale structural features, adaptively assigns weights to each branch according to the channel response value through a gating unit, and fuses and outputs a multi-scale feature map. S33, inputting the multi-scale feature map into a feature enhancement module, wherein the feature enhancement module includes a channel attention mechanism and a spatial attention mechanism, weightedly enhancing high-response channels, strengthening the spatial expression ability of the lesion area, and outputting an enhanced feature map; S34, inputting the enhanced feature map into the transformer encoding module, wherein the transformer encoding module is composed of a multi-layer stacked encoder unit, each encoder unit includes a multi-head self-attention calculation layer, a feedforward network and a residual connection structure, and introduces relative position encoding to extract context information between long-distance structures, and outputs a context encoding feature map; S35. Construct a cross-scale interactive attention module to align the scales of the encoded feature maps of different depths and establish a cross-attention map. Calculate the weighted similarity between scales through the feature matching function, fuse the contextual connections between high-expression areas and low-expression areas, and output a cross-scale fused feature map. S36: Input the cross-scale fusion feature map into the structure semantic separation unit to separate the main path features and the auxiliary path features. The main path uses the position attention mechanism, and the auxiliary path uses the context enhancement mechanism. Then, they are weighted combined through the path fusion module. S37. Perform a dimension compression mapping operation on the fused feature map, convert the multi-dimensional image features into a vector representation of uniform dimension through global average pooling and linear transformation, form the final pathological image feature vector and output it.

5. The method for pathological feature identification and negative exclusion based on microscopic imaging according to claim 1, characterized in that: The S4 specifically includes: S41, receiving the pathological image feature vector output in step S3, and expanding the feature vector into a two-dimensional spatial feature map through a linear projection layer; S42, inputting the two-dimensional spatial feature map into a multi-level classification decoding network, wherein the multi-level classification decoding network is composed of a plurality of stacked attention fusion decoding units, each of which includes an attention fusion module, a convolution restoration module, and a channel fusion module; S43. In each attention fusion module, a joint modeling mechanism of channel attention and spatial attention is introduced to dynamically enhance highly correlated feature channels through global pooling and local response adjustment strategies, and the spatial position information is used to calculate the salient area attention map; S44, inputting the attention-enhanced feature map into a context-aware fusion module, wherein the context-aware fusion module includes a local context aggregation submodule and a non-local relationship capture submodule, wherein the local context aggregation submodule calculates neighborhood semantic consistency based on a sliding window strategy, and the non-local relationship capture submodule calculates global semantic correlation between distant regions based on pixel-level feature similarity; S45. Inputting the context-aware fused feature map into a category-aware decoder, wherein the category-aware decoder includes a category embedding layer, a feature matching layer, and a category response generation layer. The category embedding layer maps the preset pathological tissue type into a high-dimensional embedding vector. The feature matching layer calculates the correlation score between each pixel or region feature and the category embedding. The category response generation layer outputs a preliminary classification probability map based on the correlation score. S46. Based on the output of the category-aware decoder, a residual enhancement path is introduced to guide the fusion of deep semantic features and shallow structural features, thereby enhancing the discrimination of classification boundaries while preserving tissue morphological details. S47. Adjust the scale consistency of the preliminary classification probability map, introduce a multi-resolution fusion strategy, upsample, normalize and weightedly fuse the output features of different spatial levels to form a multi-channel classification map of unified scale; S48, inputting the multi-channel classification map into an integrated decision module, the integrated decision module including multiple parallel classification heads, each of which performs independent discrimination based on different feature branches, and generates a final classification decision vector using a weighted method; S49: The final classification decision vector is subjected to Softmax normalization processing, and the final classification result of the pathological image is output.

6. The method for pathological feature identification and negative exclusion based on microscopic imaging according to claim 1, characterized in that: The S5 specifically includes: S51, receiving the pathological image classification determination result generated in step S4 and its corresponding pathological image feature vector, and inputting the pathological image feature vector into a feature compression mapping module, wherein the feature compression mapping module includes a fully connected layer, a normalization layer, and a dimensionality reduction transformation layer, and maps the original high-dimensional feature vector into a low-dimensional discriminant space; S52. Constructing a positive sample embedding distribution and a negative sample embedding distribution in a low-dimensional discriminant space, wherein the positive sample embedding distribution is clustered according to the characteristics of known positive cases, and the negative sample embedding distribution is constructed as a boundary template according to the characteristics of known negative cases; S53. Based on the positive and negative sample embedding distributions, construct a discriminant boundary separation model, wherein the discriminant boundary separation model includes an inter-class spacing constraint mechanism and a boundary direction enhancement mechanism to form distinguishable feature region boundaries in a low-dimensional space; S54, performing distance matching on the feature points of the image to be judged and the negative sample template in the discrimination space, using a distance weight function to evaluate the closeness to the negative distribution, and generating a discrimination similarity score; S55, inputting the discriminant similarity score into the negative probability scoring module, and generating a multi-factor evaluation vector by combining the category boundary separation strength and the discriminant confidence information; S56. Input the multi-factor evaluation vector into a negative exclusion determination module, which uses a decision tree strategy to logically combine the determination conditions, identify samples that meet the negative distribution characteristics, and output them as negative results; S57: Mark the image sample determined to be negative as "excluded" and send it to the downstream module for archiving or recording; S58. Retain the image samples determined to be non-negative.

7. The method for pathological feature identification and negative exclusion based on microscopic imaging according to claim 1, characterized in that: The S6 specifically includes: S61, receiving the negative exclusion result output in step S5 and its corresponding discriminant similarity score and multi-factor evaluation vector, and constructing a confidence vector representation for each sample; S62: Input the confidence vector into a confidence weighted evaluation module, which is composed of a boundary proximity evaluation unit, an intra-class consistency scoring unit, and a model uncertainty measurement unit, and respectively calculates the shortest boundary distance between the sample and the negative template, its local distribution consistency score within the class, and its prediction output stability index; S63. Perform weighted fusion on the three confidence score results, and generate a single comprehensive confidence score using a feature weighted aggregation mechanism. The range of the single comprehensive confidence score value is normalized to a fixed interval; S64. Construct a confidence-boundary coupling mapping model, perform bivariate modeling on the relationship between the comprehensive confidence score and the boundary position of the sample in the discriminant space, and obtain the uncertainty distribution state of the sample; S65. Dynamically adjust the threshold setting strategy based on the uncertainty distribution state and the sample category label, wherein the threshold setting strategy is based on the category density estimate and the confidence fluctuation range to determine the minimum credible judgment standard under different categories; S66: Compare the dynamic threshold with the comprehensive confidence score, mark samples below the threshold as low-confidence samples, and output them to the manual review processing path; S67, marking samples above the dynamic threshold as high-confidence samples; S68. Record all parameters and scoring results involved in the threshold determination and write them into the evaluation log file; S69. Perform confidence stratification classification on all retained samples and establish three output levels: high confidence, medium confidence and low confidence.

8. The method for pathological feature identification and negative exclusion based on microscopic imaging according to claim 1, characterized in that: The S7 specifically includes: S71, receiving the pathological image classification determination results and negative exclusion results outputted in step S4 and step S6, respectively, and generating a comprehensive information vector including a category label, determination confidence, negative status mark, and evaluation log path for each image sample; S72. Input the comprehensive information vector into a result encoding module. The result encoding module includes a field mapping unit, a format standardization unit, and a structure generation unit. The field mapping unit extracts required information based on predefined field templates, the format standardization unit standardizes various data formats to target encoding rules, and the structure generation unit creates standard structured record rows based on primary keys. S73. Writing the structured results into a pathology diagnosis database. The pathology diagnosis database adopts a multi-table joint structure. The main table records the image unique identifier and the final classification result, and the appendix records the evaluation parameters, negative judgment details, and historical recognition process metadata. S74, synchronously sending the written data to the diagnosis assistance system interface, wherein the diagnosis assistance system interface adopts a standard data transmission protocol; S75. Introduce encryption mechanisms and identity authentication processes during data transmission; S76. Generate an automatic recognition record summary for each classification result and negative exclusion mark, and push it to the doctor's work interface; S77. Set up a structured output interface log function to record and archive the time, target address, content summary, and feedback status of each result output.

Images