An Artificial Intelligence-Based Method and System for Analyzing Lung Cancer Pathological Sections

Abstract

This application belongs to the field of pathological diagnosis technology and discloses an artificial intelligence-based lung cancer pathological slide analysis and system. By introducing visualized information on the recognition basis generated by the artificial intelligence model, and combining the correction operations of multiple pathologists on the model recognition results and the acquisition of the reasons for the corrections, the consistency analysis of these correction results is further performed to generate unified data. Finally, the artificial intelligence model is incrementally trained using this data, thereby constructing a closed-loop feedback optimization mechanism that combines artificial intelligence and expert knowledge. This solves the model performance problems caused by insufficient training data coverage, expert annotation efficiency, and inconsistent results in pathological diagnosis, and realizes model learning and performance improvement.

Description

Technical Field

[0001] This application relates to the field of pathological diagnostic technology, and more specifically, to an artificial intelligence-based method and system for analyzing lung cancer pathological slides. Background Technology

[0002] In medical diagnostic practice, pathological diagnosis of lung cancer is a crucial step in determining treatment plans and assessing prognosis. Pathologists examine lung tissue sections under a microscope to identify the morphology, arrangement, and relationship of cancer cells with surrounding tissues. Accurate detection of early invasive lesions has a decisive impact on patient prognosis. However, these lesions are usually small, have indistinct morphological features, and are highly similar in morphology to carcinoma in situ or benign lesions, making identification difficult and prone to missed or misdiagnosis, posing a serious challenge to clinical diagnosis.

[0003] With the development of digital pathology technology, pathological slides are scanned into high-resolution whole-slide images (WSI), providing a data foundation for AI-assisted diagnosis. AI-based assisted diagnostic systems have been introduced into the pathology diagnostic process, utilizing deep learning models to analyze whole-slide images, aiming to automatically identify lesion areas and assist pathologists in diagnosis.

[0004] However, existing artificial intelligence models still face many challenges in practical clinical applications. Their initial training datasets have limited coverage, affecting the accuracy and reliability of diagnoses. Continuous expansion of training data and iterative model optimization are necessary to improve model performance. This requires pathologists to manually review the diagnostic results provided by the AI ​​system and accurately label cases that the system failed to identify. This new, clinically valuable data, labeled by experts, is crucial for the model to learn and understand the characteristics of complex lesions.

[0005] However, the accurate annotation process for early invasive lesions is time-consuming and highly dependent on expert experience, significantly increasing the workload of pathologists. Furthermore, for some complex cases with ambiguous morphology or unclear boundaries, subtle differences in the understanding of diagnostic criteria among different pathologists can lead to inconsistencies in annotation results. This scarcity of expert annotation resources and the subjective variability in annotation results constitute a bottleneck in acquiring high-quality training data. Due to the low efficiency and difficulty in ensuring consistency in acquiring expert-annotated data, artificial intelligence models cannot achieve rapid and effective iterative updates, resulting in a lag in the speed at which models learn and improve from clinical practice, making it difficult to adapt to constantly changing clinical needs.

[0006] Furthermore, the inherent inconsistencies in expert annotation results, even among experienced pathologists, introduce noise into the training data. When models are trained using data containing subjective differences, they may fail to learn universal characteristics of early invasive lesions and instead learn the preferences of specific experts or in specific scenarios. This further reduces the model's generalization ability and recognition accuracy when faced with real-world data from different hospitals, experts, or with different diagnostic habits, making it difficult to provide stable and reliable auxiliary diagnostic support in multi-center clinical settings. This data quality issue directly affects the ultimate clinical applicability and reliability of artificial intelligence models, urgently requiring a technical solution that can effectively address these problems.

[0007] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0008] The purpose of this application is to provide an artificial intelligence-based method and system for analyzing lung cancer pathological slides, which can improve the accuracy, reliability and clinical applicability of artificial intelligence-assisted diagnosis.

[0009] Firstly, this application provides an artificial intelligence-based method for analyzing lung cancer pathological sections, which identifies early invasive lesions in lung cancer pathological sections based on an artificial intelligence model. The steps of this method include:

[0010] A1. Obtain full-slide images of lung cancer pathology sections;

[0011] A2. Based on an artificial intelligence model, the whole slide image is analyzed to identify early invasive lesion areas and generate visualization information on the identification basis of the early invasive lesion areas;

[0012] A3. Receive correction operations performed by multiple pathologists on the early invasive lesion area based on the visualization information of the identification criteria, obtain the correction reasons corresponding to the correction operations, and obtain multiple correction results;

[0013] A4. Perform a consistency analysis on the corrected results to generate the final early invasive lesion area identification results;

[0014] A5. Based on the final early invasive lesion area identification results, the artificial intelligence model is incrementally trained.

[0015] Secondly, this application provides an artificial intelligence-based lung cancer pathological slide analysis system, which identifies early invasive lesions in lung cancer pathological slides based on an artificial intelligence model. The system includes:

[0016] The image acquisition module is used to acquire full-slide images of lung cancer pathology sections;

[0017] The image analysis module is used to analyze the whole slide image based on an artificial intelligence model, identify the early invasive lesion area, and generate visualization information on the identification basis of the early invasive lesion area.

[0018] The correction feedback module is used to receive correction operations performed by multiple pathologists on the early invasive lesion area based on the visualization information of the identification criteria, obtain the correction reasons corresponding to the correction operations, and obtain multiple correction results;

[0019] The consistency analysis module is used to perform consistency analysis on the correction results and generate the final early invasive lesion area identification result information.

[0020] The model training module is used to incrementally train the artificial intelligence model based on the final early invasive lesion area identification results.

[0021] Beneficial Effects: This application provides an AI-based lung cancer pathology slide analysis and system. By introducing visualized information on the identification criteria generated by the AI ​​model, and combining the correction operations of multiple pathologists on the model's identification results and the acquisition of the reasons for the corrections, the system further performs consistency analysis on these correction results to generate unified data. Finally, this data is used to incrementally train the AI ​​model, thereby constructing a closed-loop feedback optimization mechanism that combines AI and expert knowledge. This solves the model performance problems caused by insufficient training data coverage, expert annotation efficiency, and inconsistent results in pathological diagnosis, and realizes model learning and performance improvement. Attached Figure Description

[0022] Figure 1 A flowchart of an artificial intelligence-based lung cancer pathological slide analysis method provided in this application embodiment.

[0023] Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based lung cancer pathology slide analysis system provided in an embodiment of this application.

[0024] Labeling Explanation: 1. Image Acquisition Module; 2. Image Analysis Module; 3. Correction Feedback Module; 4. Consistency Analysis Module; 5. Model Training Module. Detailed Implementation

[0025] The technical model of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0026] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] refer to Figure 1 This application proposes an artificial intelligence-based method for analyzing lung cancer pathological sections. The method identifies early invasive lesions in lung cancer pathological sections using an artificial intelligence model. The steps of this method include:

[0028] A1. Obtain full-slide images of lung cancer pathology sections;

[0029] A2. Based on an artificial intelligence model, the whole slide image is analyzed to identify early invasive lesion areas and generate visualization information on the identification basis of the early invasive lesion areas;

[0030] A3. Receive correction operations performed by multiple pathologists on the early invasive lesion area based on the visualization information of the identification criteria, obtain the correction reasons corresponding to the correction operations, and obtain multiple correction results;

[0031] A4. Perform a consistency analysis on the corrected results to generate the final early invasive lesion area identification results;

[0032] A5. Based on the final early invasive lesion area identification results, the artificial intelligence model is incrementally trained.

[0033] Among them, artificial intelligence models refer to computational models that are trained through algorithms and are capable of analyzing image data and recognizing patterns. They can employ deep learning networks, convolutional neural networks, or recurrent neural networks, and are mainly used to automatically identify specific regions in pathological images.

[0034] Among them, the identification basis of the visualization information refers to the data that the artificial intelligence model presents in a graphical way when identifying the lesion area, which can be generated in the form of heat map, saliency map or attention map, etc. It is mainly used to show the model's judgment logic and help doctors understand and evaluate the model results.

[0035] Among them, the correction operation refers to the adjustment behavior of pathologists on the early invasive lesion area based on the recognition results and visualization information provided by the model. It can include adding, deleting, adjusting the boundary or modifying the classification of the area. It is mainly used to correct the recognition bias of the model and introduce expert experience.

[0036] Among them, the reason for correction refers to the explanatory description or classification label provided by the pathologist when performing the correction operation. It can be obtained by using preset vocabulary selection or text input. It is mainly used to record the logic of expert correction and provide semantic information for model learning.

[0037] Consistency analysis refers to the process of comparing and evaluating the revised results provided by multiple pathologists. It can be processed using voting, consensus mechanisms, or statistical methods. It is mainly used to resolve subjective differences among experts and generate unified data.

[0038] Incremental training refers to the process of continuously learning and optimizing a model based on existing models using new data. It can be achieved using techniques such as transfer learning, online learning, or reinforcement learning, and is mainly used to improve the model's ability to identify new types or complex lesions.

[0039] The core innovation of this application lies in introducing visualized information on the recognition basis generated by the artificial intelligence model, and combining the correction operations of multiple pathologists on the model's recognition results and the acquisition of the reasons for the corrections. Further consistency analysis is performed on these correction results to generate unified data. Finally, this data is used to incrementally train the artificial intelligence model, thereby constructing a closed-loop feedback optimization mechanism that combines artificial intelligence and expert knowledge. This solves the model performance problems caused by insufficient training data coverage, expert annotation efficiency, and inconsistent results in pathological diagnosis, and realizes model learning and performance improvement.

[0040] Specifically, this method establishes a closed-loop feedback system that collaborates between artificial intelligence (AI) and pathologists to continuously optimize the AI ​​model's ability to identify early invasive lesions in lung cancer pathology slides. First, the system acquires whole-slide images of lung cancer pathology slides as input data for the AI ​​model's analysis. Then, the AI ​​model analyzes these images, automatically identifying potential early invasive lesion regions and simultaneously generating visualizations of the model's basis for identifying these lesions. This visualization reveals the model's focus and decision-making logic on the images, providing a reference for human intervention. Next, the system receives correction operations performed by multiple pathologists based on these visualizations, including adding, deleting, adjusting boundaries, or modifying classifications of lesion regions. Simultaneously, the system obtains the reasons for these corrections, providing semantic information about expert judgment. By aggregating the correction operations and reasons from multiple experts, the system obtains multiple correction results. To address potential diagnostic discrepancies among experts, the system performs consistency analysis on these correction results, identifying and coordinating disagreements among different experts, ultimately generating an early invasive lesion region identification result that has undergone expert consensus or system fusion processing. Finally, using this validated and unified identification result information, the artificial intelligence model is incrementally trained, enabling the model to continuously learn and adapt from actual clinical cases and expert feedback, thereby improving its accuracy and generalization ability in identifying complex early invasive lesions.

[0041] As a preferred embodiment, the solution of this application is implemented as follows: First, digital whole-slide images of lung cancer pathological sections are acquired using a digital pathology scanner. Then, an artificial intelligence model trained on a deep convolutional neural network is used to analyze these whole-slide images, automatically detecting and marking suspected early invasive lesion areas. During the identification process, the model simultaneously generates an attention map based on gradient-weighted class activation mapping. This map is superimposed on the original image in the form of a heatmap, with color intensity indicating the model's degree of attention to different areas of the image, serving as visual information for identification. Subsequently, the system displays the model identification results and attention map to multiple pathologists through a web interface or dedicated client software. Each pathologist can interactively correct the lesion areas (abbreviated as early invasive lesion areas) identified by the model using interface tools, such as dragging the bounding box to adjust the lesion range, and selecting correction reasons from a preset list of pathological feature terms, such as "insufficient cellular atypia," "atypical acinar structure," or "unclear invasive boundaries." The system collects the correction results and corresponding correction reasons from each doctor. Next, these corrected results are compared. For example, by calculating the Jaccard similarity coefficient and classification consistency of the corrected areas (i.e., the corrected early invasive lesion areas) from different doctors, lesion areas with discrepancies are identified. For areas with discrepancies, the system can trigger an expert consultation process or generate the final lesion identification result information according to preset fusion rules. Finally, this dataset, which has undergone multi-expert correction and consistency processing, is used for incremental training of the artificial intelligence model, such as through fine-tuning or continuous learning strategies, to improve the model's performance in identifying new or difficult-to-identify lesions.

[0042] This application addresses the problems of insufficient training data coverage, inefficient expert annotation, and poor consistency in the identification of early invasive lesions in lung cancer pathological sections using artificial intelligence models. By providing visualized information on the identification basis, this solution enhances the transparency of model decision-making and assists pathologists in making efficient corrections. Simultaneously, by introducing correction operations from multiple pathologists and obtaining the reasons for these corrections, the semantic information of model learning is enriched, overcoming the limitations of single-model or single-expert judgments. Consistency analysis of the correction results eliminates data noise caused by subjective differences among experts, ensuring the quality of data used for model training. Finally, the high-quality identification results are used for incremental training of the artificial intelligence model, enabling it to continuously learn and optimize from challenging cases encountered in actual clinical practice and expert feedback. This improves the model's accuracy and generalization ability in identifying complex early invasive lesions, increases pathologists' trust in the AI-assisted diagnostic system, and enhances the efficiency and accuracy of clinical diagnosis.

[0043] In some implementations, step A2 includes:

[0044] A201. Based on the artificial intelligence model, the whole slide image is analyzed to identify the early invasive lesion area and obtain a preliminary identification image containing the boundary of the early invasive lesion area;

[0045] A202. Extract the response information of the artificial intelligence model to each region in the whole slide image during the process of identifying the early invasive lesion region;

[0046] A203. Based on the response information, generate an attention map and overlay it on the preliminary identification image to form a map-based identification image, which serves as the visual information for identifying the early invasive lesion area; the color intensity of the attention map indicates the degree of attention the artificial intelligence model pays to each region in the whole slide image.

[0047] The response information refers to the data such as the activation level, feature extraction intensity, or gradient changes of different regions of the input image at its internal layers when the artificial intelligence model processes the whole slide image. It can be realized by interpretable methods such as activation maps of intermediate layers of the model, gradient-weighted class activation mapping (Grad-CAM), or LRP (Layer-wise Relevance Propagation).

[0048] Attention maps are visual heatmaps or pseudocolor maps generated based on the response information of artificial intelligence models. They are used to intuitively represent the degree of attention or importance weight of the model in different regions of an image. They can be achieved by converting the response information into a visually recognizable intensity distribution and overlaying it onto the image.

[0049] This application aims to provide a more transparent and interpretable AI-assisted diagnostic process to improve the efficiency and quality of pathologists' corrections to model identification results. Specifically, when analyzing whole-slide images to identify early invasive lesion areas, the AI ​​model first performs preliminary image processing to identify the lesion areas and generate a preliminary identification image including their boundaries. This preliminary identification image provides pathologists with an intuitive assessment of the lesion's location and extent, forming the basis for subsequent interactions. Building upon this, to enhance the transparency of the model's decision-making, the system delves deeper into the internal operating mechanism of the AI ​​model. During the model's identification of lesion areas, the system extracts response information from various regions of the whole-slide image. This response information reflects which parts of the image had a key impact on the model's decision-making. By acquiring this internal data, this approach reveals the model's "thinking" process, providing data support for subsequent visualization interpretation. Subsequently, the system generates an attention map based on the extracted response information. This attention map visually represents the degree of attention the AI ​​model pays to different regions in the image, using varying shades of color; that is, which regions the model considers most important for identifying early invasive lesions. Overlaying this attention map onto the initial identification image creates a map-based identification image. This map-based identification image serves as a visual representation of the areas identified as early invasive lesions, not only showing the lesion regions identified by the model but, more importantly, clearly revealing the basis for the model's judgment. Pathologists reviewing the identification results can simultaneously see the key areas the model focuses on, thus gaining a more intuitive understanding of the model's decision-making logic. The provision of this map-based identification image allows pathologists to make judgments based on their understanding of the model's decision-making process when performing corrective actions. For example, if the areas the model focuses on align with key pathological features in the pathologist's experience, it increases the doctor's confidence in the model's results; conversely, if the model focuses on non-lesion areas or the focus does not match pathological features, the doctor needs to conduct a thorough review.

[0050] This transparency greatly assists pathologists in performing efficient and accurate corrections and provides high-quality reasons for corrections. This effectively improves the data quality and efficiency of subsequent incremental training of artificial intelligence models, thereby enhancing the auxiliary value and continuous improvement capabilities of artificial intelligence models in complex pathological diagnoses and solving the problems of insufficient model interpretability leading to correction difficulties and limited data quality.

[0051] Preferably, step A203 may include:

[0052] Based on the response information, determine the contribution value of each pixel region in the whole slide image to the identification result of the early invasive lesion region;

[0053] The contribution value is normalized, and the corresponding color is selected from the preset color gradient based on the normalized contribution value to generate the attention map.

[0054] The attention map is superimposed on the preliminary identification image to form a map-based identification image, which serves as the visual information for identifying the early invasive lesion area.

[0055] The response information reflects the model's focus on or feature extraction intensity for different regions of the input image. Based on this, the contribution value, after quantifying the response information, represents the magnitude of the influence of each pixel region in the whole-slide image on the final recognition result. It can be calculated based on gradients, activation maps, or specific interpretation algorithms to measure the importance of each pixel or region in the model's decision-making.

[0056] Normalization refers to the process of converting the original contribution values ​​into a uniform, preset numerical range, such as [0,1] or [0,255]. This can eliminate the differences in the units and ranges of contribution values ​​between different images or different model outputs, ensuring the consistency and comparability of subsequent color mapping.

[0057] The preset color gradient refers to a set of colors arranged in a specific order, with smooth transitions between these colors. This is used to map the normalized contribution value to visually intuitive color depth or hue changes, so that the magnitude of the contribution value can be intuitively presented through color.

[0058] This application's solution addresses the lack of precision and consistency in the generation of attention maps by introducing specific methods for quantifying, standardizing, and visualizing the response information of artificial intelligence models, thereby improving the accuracy and interpretability of the visualized information used for identification. The solution first determines the contribution value of each pixel region in the whole-slide image to the identification result of early invasive lesions based on the response information of the artificial intelligence model. This step transforms the abstract response information within the model into specific, quantifiable "contribution values," clarifying which regions in the image play a key role in the model's identification decisions, thus providing a precise quantitative basis for subsequent attention map generation and solving the problem of response information being difficult to directly convert into visualization evidence. Next, the contribution values ​​are normalized, and based on the normalized contribution values, corresponding colors are selected from a preset color gradient to generate the attention map. By normalizing the contribution values, it ensures that the contribution values ​​output by different cases or different models can be compared and mapped on a unified scale, eliminating potential dimensional differences in the original contribution values ​​and guaranteeing the consistency of the visualization results. Based on this, corresponding colors are selected from a preset color gradient according to the normalized contribution values, providing a standardized mapping rule for the generation of the attention map. This preset color gradient establishes a clear and intuitive correspondence between color intensity and contribution value, enabling pathologists to quickly and accurately understand the degree of attention or contribution of the artificial intelligence model to different regions, thereby improving the interpretability and practicality of the attention map. Finally, the generated attention map is superimposed on the preliminary recognition image to form a map-based recognition image, serving as visual information for the identification of early invasive lesion areas.

[0059] By overlaying precisely quantified and standardized attention maps onto preliminary identification images, pathologists can intuitively see the basis for the AI ​​model's identification judgments—namely, the key areas the model focused on and their degree of contribution—while reviewing the identification results. This atlas-based identification image provides pathologists with transparent and traceable diagnostic assistance information, helping them quickly understand the model's decision-making logic and efficiently perform manual review and correction.

[0060] As a preferred implementation, when determining the contribution of each pixel region in the whole slide image to the identification result of early invasive lesion regions, a gradient-weighted class method, such as Grad-CAM (Gradient-weighted Class Activation Mapping) or LRP (Layer-wise Relevance Propagation) algorithm, can be used. Specifically, the gradient of the predicted score of the target category (early invasive lesion) relative to a specific feature map (e.g., if a convolutional neural network is used, feature maps output by the model at specific convolutional or fully connected layers can be extracted, reflecting the activation intensity of the model for different regions of the image) can be calculated, and these gradients can be used as weights to perform a weighted summation of the feature maps, thereby obtaining an original heatmap representing the degree of contribution of each pixel region. When normalizing the contribution values, the original heatmap obtained above can be subjected to min-max normalization, for example, scaling all pixel values ​​in the heatmap to the range of 0 to 255, where 0 represents the lowest contribution and 255 represents the highest contribution. When selecting the corresponding color from a preset color gradient based on the normalized contribution value, a color gradient from transparent (or low-saturation blue) to high-saturation red (or yellow) can be used. Areas with lower contribution values ​​can be mapped to transparent or light blue, while areas with higher contribution values ​​can be mapped to dark red or bright yellow, visually representing the increasing degree of attention. This generates an attention map that matches the size of the entire slide image. When superimposing the attention map onto the preliminary identification image, image fusion techniques, such as alpha blending, can be used to superimpose the generated attention map onto the preliminary identification image in a semi-transparent manner. For example, the transparency of the attention map can be set to 0.5, so that it clearly displays the area of ​​interest without completely obscuring the details of the original pathological image.

[0061] In some implementations, step A3 includes:

[0062] A301. Display the identification criteria visualization information to the pathologist as reference information for corrective actions on the early invasive lesion area;

[0063] A302. Receive the correction operation performed by the pathologist on the early invasive lesion area; the correction operation includes at least one of adding, deleting, adjusting the boundary and modifying the classification of the early invasive lesion area;

[0064] A303. Based on the type of the correction operation and the image features of the corresponding correction area, extract a list of candidate correction reasons related to the correction operation from a preset pathological feature vocabulary library and display it to the pathologist;

[0065] A304. Receive the correction reason selected by the pathologist from the candidate correction reason list, and / or receive the supplementary correction reason input by the pathologist;

[0066] A305. Normalize the selected correction reasons and / or the input supplementary correction reasons to obtain structured correction reasons;

[0067] A306. Associate the structured correction reason with the correction operation and record it to obtain the correction result.

[0068] The pre-built pathological feature vocabulary refers to a pre-constructed collection of standard pathological terms and concepts, which can be organized in the form of medical ontology, controlled vocabulary, or professional domain knowledge graph. For example, it can include terms such as "benign lesion", "inflammatory response", "fibrosis", "necrosis", "cellular atypia", and "insufficient depth of invasion". These terms have been reviewed and classified by experts to provide accurate and consistent options for correcting causes.

[0069] Normalization refers to the process of converting unstructured or semi-structured text information into a unified, machine-readable structured format. It can be achieved using natural language processing techniques (such as named entity recognition and semantic analysis), rule matching, or ontology-based mapping methods. For example, "the lesion is benign" can be normalized to "benign lesion", or "unclear boundaries" can be normalized to "blurred boundaries".

[0070] Among them, the structured correction reason refers to the correction reason represented in a predefined data format (e.g., key-value pairs, JSON objects, or database records) after normalization processing. The correction result obtained by associating the correction operation with the record can include the correction type, the image feature label of the correction area, and the corresponding standard pathological reason code or text. For example, it can be represented as {"Operation type":"Delete", "Image feature":"Inflammation", "Cause":"Benign inflammatory lesion"}, which facilitates the system to store, retrieve, and analyze the correction reason.

[0071] This solution features a meticulously designed process for pathologists to correct procedures and identify the reasons for those corrections, creating a highly efficient and structured feedback loop. First, before a pathologist performs any corrective actions on an early invasive lesion area, the system displays visualized information illustrating the basis for the identification. This allows the pathologist to intuitively understand the underlying logic and focus of the AI ​​model's judgment, enabling them to more accurately and quickly pinpoint model errors. Using this information as a reference, they can then perform corrective actions such as adding, deleting, adjusting boundaries, or modifying classifications. This transparent information presentation enhances the accuracy and efficiency of the pathologist's corrective actions.

[0072] Upon receiving a pathologist's correction request, the system doesn't simply wait for free text input. Instead, based on the specific type of correction and the image features of the area being corrected, it intelligently filters and extracts a list of candidate correction reasons relevant to the current context from a pre-defined pathological feature vocabulary. This mechanism reduces the burden on pathologists to think about and input correction reasons while ensuring the professionalism and standardization of the provided options. Pathologists can quickly select the appropriate reason from this intelligently generated list, which not only improves operational efficiency but also effectively avoids inconsistencies in reason descriptions caused by differences in personal expression habits among different doctors.

[0073] To accommodate special cases and deeper expert insights, the system provides standard options while also allowing pathologists to input supplementary correction reasons. Regardless of whether the correction reasons are selected or supplemented, all received reasons are further standardized, transforming them into uniform, machine-readable structured reasons. This process eliminates linguistic ambiguity and differences in expression, ensuring that all expert feedback data can be accurately understood and utilized by the system. Ultimately, these structured correction reasons are linked to the specific correction actions performed by the pathologist, forming a complete correction result.

[0074] The coordinated operation of this series of steps enables the AI ​​model to receive high-quality, valuable supervisory data. By providing visualized information to support identification, pathologists can make corrections more effectively, while a guided, structured cause-finding process ensures the standardization and consistency of the correction reasons. These meticulously processed correction results provide a foundation for subsequent targeted incremental training of the AI ​​model. For example, when the model misses a specific type of lesion, analyzing the structured correction reasons can clarify whether it is due to specific causes such as "insufficient cellular atypia" or "interference from background inflammation," thus guiding the model to focus more on learning these subtle features during incremental training, improving the model's generalization ability and identification accuracy. This mechanism allows the AI ​​model to continuously learn and improve from clinical practice, forming a self-optimizing closed loop, enhancing the practicality and reliability of the assisted diagnostic system.

[0075] Preferably, step A303 may include:

[0076] Based on the type of the correction operation and the image features of the corresponding correction area, a preliminary list of candidate correction reasons related to the correction operation is extracted from a preset pathological feature vocabulary.

[0077] Based on the image features of the correction area, the correlation between each cause in the preliminary extracted list of candidate correction causes and the image features is analyzed to obtain the feature correlation degree of each candidate correction cause.

[0078] Obtain the co-occurrence frequency information of the correction reasons for early invasive lesion regions with image features similar to the correction region in historical correction data;

[0079] Combining the feature correlation degree and the co-occurrence frequency information, the preliminary extracted candidate correction reason list is prioritized or grouped to obtain an optimized candidate correction reason list;

[0080] The optimized list of candidate corrective reasons is displayed to the pathologist.

[0081] The image features of the correction area refer to the visual information presented by the early invasive lesion area when the pathologist performs correction operations, such as cell morphology, tissue structure, staining intensity, clarity of lesion boundaries, nucleus-cytoplasm ratio, degree of atypia, and invasion pattern. These features can be acquired and quantified using image processing techniques, such as deep learning-based feature extractors, texture analysis algorithms, and color histogram analysis.

[0082] Feature correlation refers to the degree of matching or correlation between the initially extracted candidate causes for correction and the image features of the correction region. It can be calculated in various ways, such as by evaluating the semantic difference between the cause description and the image feature description through a pre-trained semantic similarity calculation model, or by learning the mapping relationship between image features and the cause for correction through a machine learning model and outputting a correlation score.

[0083] Co-occurrence frequency information refers to the statistical frequency or probability in which lesion regions with similar image features to the currently corrected region are selected for a specific correction reason in historical correction data. It can be obtained using statistical analysis methods, such as constructing a co-occurrence matrix, calculating conditional probabilities, or using association rule mining algorithms to quantify the degree of co-occurrence between specific image features and correction reasons.

[0084] After receiving a pathologist's correction operation on an early invasive lesion area, this solution, to improve the efficiency of obtaining the correction reasons, firstly extracts a preliminary list of candidate correction reasons related to the correction operation from a pre-defined pathological feature vocabulary based on the type of correction operation and the corresponding image features of the correction area. This preliminary extraction ensures that the scope of the initial list is roughly related to the current correction behavior and image content, laying the foundation for subsequent fine processing. Based on this, the system further analyzes the correlation between each reason in the preliminary extracted candidate correction reason list and the image features, thereby obtaining the feature correlation degree of each candidate correction reason. This analysis step delves into the visual information of the correction area, quantifying the matching degree between each candidate reason and the current image features, enabling the system to identify which reasons semantically best match the specific image manifestation observed by the pathologist. This provides an objective basis for subsequent optimization based on image content. Simultaneously, to incorporate the accumulated experience and historical patterns of the expert group, the system obtains the co-occurrence frequency information of correction reasons for early invasive lesion areas with similar image features to the correction area in historical correction data. By analyzing a large amount of historical correction data, the system can understand which correction reasons are frequently selected by pathologists in cases with image features similar to the current correction area. This reflects common problems in clinical practice, diagnostic consensus, or typical correction reasons corresponding to specific pathological features, providing an experience-based statistical foundation for optimization. Subsequently, this approach combines feature correlation and co-occurrence frequency information to prioritize or group the initially extracted candidate correction reason list, resulting in an optimized candidate correction reason list. This is the core mechanism of this approach. By comprehensively considering the visual relevance of the reason to the current image and the frequency with which the reason is selected in similar historical cases, the system can rearrange or classify candidate reasons according to an intelligent strategy. For example, correction reasons that are both highly correlated with the current image features and frequently appear in historical data will be given higher priority or placed in a more prominent position. This two-dimensional optimization ensures that the list presented to pathologists is both visually accurate and conforms to the experience patterns of clinical practice, significantly improving the possibility for pathologists to quickly locate and select accurate correction reasons. Finally, the system displays the optimized candidate correction reason list to the pathologist. By presenting a smartly filtered and sorted list, pathologists no longer need to spend a lot of time searching through redundant or unrelated options. They can directly select from a highly relevant and easy-to-understand list. This coordinated process enables pathologists to provide corrective feedback more efficiently and accurately, thus providing more valuable and consistent structured data for the continuous learning and improvement of artificial intelligence models.Compared to the basic approach of only initially extracting a list, this significantly improves the value and efficiency of data collection, thereby enhancing the ability of AI models to perform effective incremental training through this feedback and solving the problem of insufficient feedback data quality limiting model performance improvement.

[0085] In one specific embodiment, when a pathologist performs a correction operation on an early invasive lesion area identified by an artificial intelligence model—for example, correcting an area marked as benign by the model to an early invasive lesion—the system first identifies the type of correction operation as "classification modification" and acquires the image features of the modified area, such as at least one of its cell nuclear size, mitotic figures, glandular structure, and invasion depth. Based on this information, the system initially filters candidate correction reasons related to "classification modification" and "early invasive lesion" from a pre-set pathological feature vocabulary, such as "invasive adenocarcinoma," "micropapillary structure," and "atypical adherent growth pattern." Next, based on the image features of the modified area, the system uses a pre-trained image feature analysis module to evaluate the degree of matching between each initially extracted candidate correction reason and the current image features. For example, if the modified area has a micropapillary structure, then the feature correlation of the reason "micropapillary structure" will be calculated as a high value. This correlation can be a value between 0 and 1, representing the degree of semantic and visual matching. Simultaneously, the system queries a database containing a large number of historical correction records. This database records past correction operations performed by pathologists on early invasive lesion areas with similar image features, along with the corresponding reasons for these corrections. The system uses an image feature similarity matching algorithm to identify historical cases with image features highly similar to the currently corrected area and calculates the co-occurrence frequency of each selected correction reason in these similar cases. For example, if "micropapillary structure" was selected frequently in the historical data for lesion corrections, its co-occurrence frequency information will be recorded. Subsequently, the system combines the calculated feature correlation with the obtained co-occurrence frequency information. For example, a weighted summation method can be used to fuse the feature correlation and co-occurrence frequency information to generate a comprehensive evaluation value. Correction reasons with higher comprehensive evaluation values ​​will be given higher priority. For example, if "micropapillary structure" has both a high feature correlation and a high co-occurrence frequency, it will be placed at the top of the list. If a cause has high feature correlation but low co-occurrence frequency (usually representing a rare case), or low feature correlation but high co-occurrence frequency (usually representing a common but non-specific cause), the system will balance them according to preset weighting rules. Finally, the system will prioritize the initially extracted candidate cause list based on these comprehensive evaluation values, forming an optimized list. This optimized candidate cause list will then be displayed to the pathologist through the user interface. The pathologist can quickly select a cause that highly matches the current correction from this intelligently sorted list, for example, by directly clicking "micropapillary structures," thus completing the feedback for the corrected cause.

[0086] Preferably, the step of prioritizing or grouping the initially extracted candidate correction reason list by combining the feature correlation degree and the co-occurrence frequency information to obtain an optimized candidate correction reason list may include:

[0087] Determine the statistical sparsity of the image features in the corrected region;

[0088] Based on the statistical sparsity, the contribution weights of the feature correlation degree and the co-occurrence frequency information in generating the optimized candidate correction reason list are determined;

[0089] Based on the contribution weight, combined with the feature correlation degree and the co-occurrence frequency information, the initially extracted candidate correction reason list is prioritized or grouped to obtain the optimized candidate correction reason list.

[0090] Statistical sparsity refers to the degree to which the image features of the corrected region appear less frequently in historical corrected data. It can be achieved by calculating the number of times the image features appear in the historical dataset and comparing them with a preset threshold, or by analyzing their percentile ranking in the feature distribution.

[0091] The contribution weight refers to the relative importance or influence of feature correlation and co-occurrence frequency information when generating the optimized candidate correction reason list. It can be dynamically adjusted using adaptive algorithms, machine learning models, or based on preset rules to ensure that the reliability differences of different information sources are reflected.

[0092] During priority sorting or grouping, each reason in the initially extracted candidate correction reason list can be weighted and scored according to a determined contribution weight. For example, the final score of each candidate correction reason can be calculated as: Score = (Feature correlation degree * Feature correlation degree weight) + (Co-occurrence frequency information * Co-occurrence frequency information weight). The system can sort the candidate correction reasons in descending order based on these scores, or group them according to the score falling into a preset score range (for example, if the score falls into the preset high score range, the corresponding candidate correction reason is assigned to the "Highly Recommended" group; if the score falls into the preset medium score range, the corresponding candidate correction reason is assigned to the "Moderately Recommended" group; if the score falls into the preset low score range, the corresponding candidate correction reason is assigned to the "Generally Recommended" group), thereby obtaining an optimized candidate correction reason list, which is then displayed to the pathologist.

[0093] This method first determines the statistical sparsity of image features in the correction region based on co-occurrence frequency information. This step assesses the statistical prevalence or rarity of image features in the current correction region by analyzing their frequency of occurrence in historical data. Image features with low frequency in historical data, i.e., high statistical sparsity, indicate that their corresponding historical co-occurrence frequency information may be insufficient or unrepresentative, thus providing a basis for subsequent weight allocation. Next, based on the determined statistical sparsity, the contribution weights of feature correlation and co-occurrence frequency information in generating the optimized candidate correction cause list are dynamically determined. This means that the system no longer simply combines the two types of information in a fixed ratio, but adjusts their relative importance according to the characteristics of the current case. For example, when the statistical sparsity of image features in the correction region is high, the system assigns higher weight to the feature correlation obtained based on the analysis of current image features, because historical co-occurrence frequency information may not be sufficient to provide a reliable reference; conversely, for common cases with low statistical sparsity, historical co-occurrence frequency information is assigned higher weight to fully utilize its statistical reliability. This adaptive weighting mechanism enables the system to more intelligently utilize information from different sources, avoiding the limitations of a single information source that could affect the accuracy of the results. Finally, based on the determined contribution weights, combined with feature correlation and co-occurrence frequency information, the initially extracted candidate correction reason list is prioritized or grouped to obtain an optimized candidate correction reason list. Through this weighted approach, the system can more accurately sort or group candidate correction reasons, ensuring that the correction reason list presented to pathologists not only includes options highly relevant to the current image features but also considers the accumulation of historical experience and can intelligently adjust based on the rarity of cases.

[0094] This approach, combined with previous techniques, further enhances the performance of the entire lung cancer pathology slide analysis method. Specifically, in the step of receiving corrections from pathologists and obtaining the reasons for those corrections, this approach dynamically adjusts the combination of feature correlation and co-occurrence frequency information, resulting in a more accurate optimization of the initially extracted list of candidate correction reasons. This optimization directly affects the list of candidate correction reasons displayed to the pathologist, providing them with more targeted and reliable reference information during correction operations. This not only improves the efficiency of pathologists in selecting correction reasons but also ensures the accuracy of the selected reasons, especially when dealing with rare or atypical lesions, effectively avoiding misjudgments due to insufficient historical data. Ultimately, these high-quality correction results will be used for incremental training of the artificial intelligence model, enabling the model to learn from more accurate and representative expert feedback, continuously improving its ability to identify and generalize early invasive lesions, forming an efficient and adaptive closed-loop optimization system.

[0095] In some implementations, step A4 includes:

[0096] A401. Compare the boundary information and classification information of each early invasive lesion region in the multiple correction results, identify the early invasive lesion regions with correction differences, and determine the type of correction difference;

[0097] A402. Based on the type of correction difference, extract the structured correction reasons corresponding to the early invasive lesion areas with correction differences;

[0098] A403. Based on the modified difference type and the extracted structured modification reasons, assess the clinical significance of the modified difference;

[0099] A404. Based on the clinical significance of the correction difference, display to the multiple pathologists the areas of early invasive lesions with correction differences and the corresponding extracted structured correction reasons to assist the pathologists in their discussions;

[0100] A405. Receive the consensus opinion formed after discussion by the pathologists, or according to the preset fusion rules, combine the clinical importance of the correction differences, and fuse the multiple correction results to generate the final early invasive lesion area identification result information.

[0101] Among them, correction discrepancies refer to inconsistencies between different pathologists' correction operations (such as boundary adjustment, classification modification, addition, or deletion) on the same early invasive lesion area. Correction discrepancies can manifest as insufficient boundary overlap, inconsistent classification labels, or a lesion being added by some doctors while others do not. They can be addressed using image processing algorithms (such as IoU calculation and pixel-level difference analysis) and text semantic analysis (such as classification label comparison).

[0102] The clinical significance of the corrected difference refers to the quantitative assessment of the potential clinical impact of differences in the corrected results among different pathologists. It reflects the criticality of the difference in patient diagnosis, treatment selection, or prognosis. This can be achieved through methods such as expert scoring, weighted allocation based on clinical guidelines, or prediction using machine learning models.

[0103] The pre-defined fusion rules refer to the system's strategy for integrating multiple corrected results according to pre-set logic when pathologists fail to reach a consensus or when automated processing is required. For example, weighted fusion can be based on the clinical importance of the corrected differences, the pathologist's qualifications, the pathologist's historical diagnostic accuracy, or majority voting principles. This can be implemented using algorithms such as weighted averaging, Bayesian fusion, decision trees, or rule engines.

[0104] The operational logic of this solution is as follows: First, after receiving multiple correction operations performed by pathologists on early invasive lesion areas and obtaining the reasons for these corrections, resulting in multiple correction results, the system compares the boundary and classification information of each early invasive lesion area in these results to identify correction areas where there are differences in corrections among different pathologists, and further determines the specific types of these differences. This comparison process is the foundation for subsequent processing; it clarifies the inconsistencies that need attention and resolution, avoids blindly processing all results, and improves processing efficiency and targeting. Next, based on the identified types of correction differences, the system can extract the structured correction reasons corresponding to the early invasive lesion areas with correction differences. This step is crucial because it links the correction operation to the fundamental reason why the pathologist made that correction. By obtaining the structured correction reasons, the system not only knows "where the difference is," but also "why the difference is," providing important semantic information for subsequent difference assessment and resolution. This is because when obtaining the correction operations, the system has already received and standardized the correction reasons selected or input by the pathologists, giving these reasons a structured form that can be analyzed and utilized by the system. Subsequently, based on the corrected difference type and the extracted structured corrected reasons, the system can assess the clinical significance of the corrected differences. This assessment step is one of the key innovations of this protocol. It goes beyond simple difference detection, quantifying the potential impact of differences from a clinical perspective. For example, critical differences that may affect patient treatment plans or prognosis will have higher clinical significance. This assessment helps distinguish between "unimportant minor differences" and "critical differences that need to be addressed," ensuring that resources are focused on resolving valuable differences. Then, based on the assessed clinical significance of the corrected differences, the system displays the areas of early invasive lesions with corrected differences and the corresponding extracted structured corrected reasons to multiple pathologists to assist in discussions. This step demonstrates the advantages of human-computer collaboration. The system identifies and quantifies the differences and provides contextual information (corrected reasons), which pathologists can use for targeted discussions, leveraging their expertise and experience to reach a consensus. This guided discussion mechanism improves the efficiency and quality of expert consensus formation. Finally, the system receives the consensus reached by pathologists after discussion, or, based on preset fusion rules and the clinical importance of the differences, fuses multiple corrected results to generate the final early invasive lesion area identification result. This step is the endpoint of the entire consistency analysis process, ensuring that a single, high-quality identification result with clinical consensus or intelligent fusion is ultimately produced from multiple potentially inconsistent corrected results.Whether through expert discussions to reach a consensus or through automated processing via intelligent fusion rules, the clinical significance of the differences is fully considered, thereby ensuring the accuracy and reliability of the final results and providing a solid foundation for the continuous learning and optimization of artificial intelligence models.

[0105] Overall, this approach effectively addresses the shortcomings of existing technologies in terms of poor consistency among multi-expert labeled data by introducing a consistency analysis mechanism for multi-source correction results and handling differences based on the reasons for correction and clinical importance. Through systematic difference identification, cause extraction, importance assessment, expert-assisted discussion, and intelligent fusion, this approach transforms previously scattered and potentially conflicting expert correction opinions into unified and high-quality final identification results. This not only solves the problem of decreased model training performance due to data inconsistency, but more importantly, it integrates the professional experience and judgment of pathologists into the data refinement process, making the final generated data more clinically valuable and reliable. This provides a solid foundation for the continuous incremental training of the artificial intelligence model, improves the model's generalization ability and recognition accuracy in complex clinical scenarios, and enhances its practicality and credibility in multi-center clinical environments.

[0106] In one specific embodiment, this solution can be applied to processing the correction results of three pathologists (e.g., Doctor A, Doctor B, and Doctor C) for the same early invasive lesion region identified by an artificial intelligence model. First, in the comparison phase, the system receives the correction results submitted by Doctor A, Doctor B, and Doctor C respectively. Assume Doctor A labels the lesion region as "adenocarcinoma" and outlines its perceived boundary; Doctor B also labels it as "adenocarcinoma," but its outlined boundary partially overlaps with Doctor A's but also differs; while Doctor C labels the lesion region as "atypical adenomatous hyperplasia," and its boundary differs from both of the previous two doctors. The system automatically compares the boundary information (e.g., by calculating pixel-level differences or IoU values) and classification information (e.g., by directly comparing classification labels) of the lesion region in these three correction results. The comparison results show that the lesion region exhibits two types of correction discrepancies: "inconsistent boundary" and "inconsistent classification." Next, based on the identified correction discrepancy types, the system automatically extracts the structured correction reasons associated with the correction results submitted by Doctors A, B, and C. For example, Doctor A's boundary correction reason is "insignificant cellular infiltration at the lesion margin," Doctor B's is "interference from local inflammatory response," and Doctor C's classification correction reason is "insufficient cellular atypia, not meeting the diagnostic criteria for adenocarcinoma." These structured reasons provide the underlying basis for the differences. Subsequently, the system assesses the clinical importance of these differences based on the types of corrections and the extracted structured correction reasons. For example, because there are significant differences in clinical treatment between "adenocarcinoma" and "atypical adenomatous hyperplasia," the system will assess "classification inconsistency" as having high clinical importance; while minor differences in the boundary, if they do not affect the overall assessment of the lesion's nature, may be assessed as having moderate or low importance. This assessment can be combined with the lesion's pathological characteristics (e.g., size, location, cell density) and the patient's clinical background information (e.g., age, medical history). Then, based on the assessed clinical importance, the system displays images of early invasive lesion areas with correction discrepancies, the correction results of Doctor A, Doctor B, and Doctor C, and the corresponding structured reasons for correction on a single interactive interface, highlighting discrepancies with high clinical importance. For example, the system specifically indicates that "classification inconsistency" is a high-importance discrepancy and displays Doctor C's correction reason of "insufficient cellular atypia." This assists the three pathologists in conducting targeted discussions. For instance, they can jointly review the original slide images and, based on their respective professional experience, conduct a detailed analysis of the lesion's cell morphology and tissue structure. Finally, after the discussion, the system receives the consensus reached by the pathologists. For example, after discussion, the three doctors ultimately reach a consensus that the lesion should be diagnosed as "atypical adenomatous hyperplasia" and jointly determine a final boundary. The system uses this consensus as the final early invasive lesion area identification result.Alternatively, if pathologists fail to reach a clear consensus, the system can also automate the process according to pre-defined fusion rules. For example, for classification differences of high clinical importance, the system may prioritize the opinion of the majority of experts; for boundary differences, the system can generate a comprehensive boundary based on its clinical importance using a weighted average or an intelligent fusion algorithm based on morphological features, thereby obtaining the final identification result of early invasive lesion areas.

[0107] Preferably, step A403 may include:

[0108] Obtain pathological features of early invasive lesion areas with correction differences, as well as the corresponding patients' clinical background information;

[0109] Based on the pathological features and clinical background information, the weights of the assessment indicators for the corrected differences in different clinical dimensions are determined from a pre-defined clinical assessment rule base.

[0110] The clinical significance of the modified difference is calculated based on the modified difference type, the structured modification reason, and the assessment index weight.

[0111] Among them, pathological feature information refers to data describing the morphological, histological and cytological characteristics of early invasive lesion areas. It can be represented by data from multiple dimensions, such as lesion size, shape, degree of cellular atypia, number of mitotic figures, depth of invasion, vascular invasion, distribution of necrotic areas and immunohistochemical staining results.

[0112] Clinical background information refers to data related to a patient's individual health status, disease progression, and treatment. It can be represented by various data types, including the patient's age, gender, medical history (such as smoking history and family history of cancer), comorbidities, imaging results, blood biochemical indicators, and the treatment regimens received.

[0113] The pre-defined clinical assessment rule base refers to a knowledge system that includes expert knowledge, clinical guidelines, and historical case data. It can be implemented in various forms, such as rule-based expert systems, decision tree models, Bayesian networks, or machine learning models. It is used to derive the importance assessment results (i.e., assessment indicator weights) of the corrected differences in different clinical dimensions based on the input pathological feature information and clinical background information.

[0114] Among them, the weight of the assessment index refers to the relative importance coefficient assigned to different clinical dimensions (such as the impact on diagnostic accuracy, the impact on treatment selection, and the impact on patient prognosis) when assessing the clinical importance of the corrected difference. It can be expressed in numerical, percentage, or rank form, reflecting the degree of contribution of each factor to the final assessment result under specific pathological and clinical situations.

[0115] Among them, comprehensive calculation refers to the integration and processing of multi-source information such as the type of correction difference, the reasons for structured correction, and the weight of assessment indicators to obtain a unified and quantitative clinical importance assessment result. It can be achieved by various algorithms such as weighted summation, multivariate regression analysis, or machine learning-based fusion models.

[0116] This method, building upon existing assessments based on modified difference types and structured causes of modification, further incorporates pathological features of early invasive lesion areas and patient clinical background information. Specifically, firstly, the system acquires pathological features of early invasive lesion areas exhibiting modified differences. This information directly reflects the lesion's biological behavior and potential malignancy, such as size, morphology, and cellular atypia. Simultaneously, the system acquires the corresponding patient's clinical background information, providing individualized risk factors and prognostic considerations, such as age, medical history, and treatment regimen. The introduction of this data allows the assessment of modified differences to move beyond the type and superficial causes of the differences, delving into the actual situation of the lesion and the patient, thus more accurately determining the true clinical significance of the difference. Based on this, the system determines the weights of assessment indicators for modified differences across different clinical dimensions from a pre-defined clinical assessment rule base, using the acquired pathological features and clinical background information. This process achieves intelligent and personalized assessment, as different pathological features and clinical backgrounds can lead to drastically different clinical importance for the same type of modified difference. By utilizing a pre-defined rule base and considering specific case characteristics, the system dynamically assigns corresponding weights to various assessment indicators (such as their impact on diagnosis, treatment, and prognosis), ensuring the clinical relevance and specificity of the assessment results. Finally, the system comprehensively calculates the clinical importance of the corrected differences based on the type of correction, the structured reasons for correction, and the dynamically determined weights of the assessment indicators. This step integrates all relevant information to arrive at a comprehensive and instructive assessment of clinical importance. By combining the type of corrected differences (e.g., boundary differences, classification differences), the structured reasons for correction provided by the pathologist (e.g., "inconsistent judgment on the proportion of invasive adenocarcinoma components"), and the weights of the assessment indicators determined based on pathological and clinical information, the system can perform multi-dimensional, weighted calculations. This comprehensive calculation method avoids the limitations of single-factor assessments, ensuring that the assessment results fully reflect the potential impact of corrected differences in actual clinical decision-making.

[0117] Through the coordinated operation of the above steps, this method expands the assessment dimensions from the single correction difference itself to the intrinsic characteristics of the lesion and the individual patient's condition, making the assessment results more clinically valuable. This more refined and personalized assessment can effectively distinguish the true impact of different correction differences in actual clinical diagnosis and treatment, thus providing a solid and reliable basis for subsequent discussions among pathologists, reaching consensus, or automatic fusion of results by the system, significantly improving the accuracy of the final identification results.

[0118] Preferably, the step of comprehensively calculating the clinical significance of the modified difference based on the modified difference type, the structured modification reason, and the assessment index weight may include:

[0119] The types of correction differences and the reasons for the structured corrections are converted into corresponding quantitative importance values, respectively.

[0120] The quantitative importance value is weighted according to the weight of the evaluation index;

[0121] The clinical importance of the corrected difference is obtained by aggregating and weighting the quantitative importance values.

[0122] Among them, quantitative importance value refers to converting the originally non-numerical correction difference type and structured correction reason into a numerically meaningful indicator through preset mapping rules, scoring system or machine learning model. It can be achieved by numerical scoring, level mapping or probability assignment.

[0123] Weighted processing refers to applying different weight coefficients to the quantitative importance value according to different evaluation dimensions to reflect its relative contribution in the overall evaluation. It can be achieved by multiplicative weighting, proportional allocation or statistical model-based methods.

[0124] Aggregation refers to combining multiple weighted quantitative importance values ​​into a single, comprehensive value through specific mathematical operations or logical rules to represent the overall clinical importance of the corrected difference. This can be achieved using summation, average calculation, or a multi-factor comprehensive evaluation model.

[0125] This approach aims to provide an objective and standardized method for comprehensively calculating the clinical importance of corrected differences, addressing the challenge of quantifying and effectively aggregating qualitative information during the assessment process, thereby improving the accuracy and reliability of the evaluation. Specifically, the method first converts the type of corrected difference and the structured reasons for correction into quantitative importance values. This step solves the problem of transforming qualitative or semi-qualitative information into calculable values, laying the foundation for subsequent comprehensive calculations and ensuring the objectivity of the assessment. Subsequently, the quantitative importance values ​​are weighted according to the weights of the assessment indicators. These weights reflect the importance or impact of corrected differences across different clinical dimensions. This weighting ensures that the relative importance of different clinical dimensions is fully considered when calculating clinical importance, making the final assessment results more aligned with actual clinical needs. Finally, the weighted quantitative importance values ​​are aggregated to obtain a single, comprehensive value representing the clinical importance of the corrected difference. This aggregation method integrates previously scattered, multi-dimensional information into a unified indicator, facilitating pathologists' rapid understanding and comparison of the severity of different corrected differences. The implementation of this scheme has shifted the assessment of correction differences in early invasive lesion areas from qualitative judgment to quantitative analysis, providing objective and quantitative evidence for subsequent pathologist discussions, consensus formation, or automated fusion processing, thereby improving the efficiency and accuracy of the entire analysis process.

[0126] In one implementation, when converting the types of corrections and the reasons for structured corrections into corresponding quantitative importance values, an initial importance score can be preset for each type of correction (e.g., "boundary adjustment," "classification modification," "addition," "deletion") and stored in a lookup table. For example, "boundary adjustment" can be assigned 3 points, and "classification modification" can be assigned 5 points. Similarly, for the reasons for structured corrections (e.g., "atypical lesions," "blurred boundaries," "cellular atypia"), corresponding scores can also be preset, such as "atypical lesions" at 4 points and "blurred boundaries" at 2 points. These scores can be determined based on clinical expert consensus or historical data analysis. When weighting the quantitative importance values ​​according to the assessment indicator weights, the assessment indicator weights for different clinical dimensions can be obtained first, as described above. Then, the previously obtained quantitative importance values ​​are multiplied by the corresponding assessment indicator weights to obtain the weighted values ​​(each quantitative importance value corresponds to the assessment indicator weight for each clinical dimension, resulting in a corresponding weighted quantitative importance value). When aggregating the weighted quantitative importance values, all weighted quantitative importance values ​​can be simply summed, or a weighted average can be used to calculate a final comprehensive value for the clinical importance of the corrected difference. For example, the final clinical importance can be expressed as the sum of (the quantitative value of the corrected difference type multiplied by its corresponding weight) and (the quantitative value of the structured corrected cause multiplied by its corresponding weight).

[0127] refer to Figure 2 This application provides an artificial intelligence-based lung cancer pathology slide analysis system, which identifies early invasive lesions in lung cancer pathology slides based on an artificial intelligence model. The system includes:

[0128] Image acquisition module 1 is used to acquire whole-slide images of lung cancer pathological sections (for details, please refer to step A1 above).

[0129] Image analysis module 2 is used to analyze the whole slide image based on an artificial intelligence model, identify the early invasive lesion area, and generate visualization information on the identification basis of the early invasive lesion area (the specific process can be referred to step A2 above).

[0130] The correction feedback module 3 is used to receive correction operations performed by multiple pathologists on the early invasive lesion area based on the visualization information of the identification criteria, and to obtain the correction reasons corresponding to the correction operations, and to obtain multiple correction results (for details, please refer to step A3 above).

[0131] Consistency analysis module 4 is used to perform consistency analysis on the correction results and generate the final early invasive lesion area identification result information (for details, please refer to step A4 above).

[0132] Model training module 5 is used to incrementally train the artificial intelligence model based on the final early invasive lesion area identification results (the specific process can be referred to step A5 above).

[0133] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An artificial intelligence-based method for analyzing lung cancer pathological sections, characterized by identifying early invasive lesions in lung cancer pathological sections using an artificial intelligence model, wherein... The steps of this method include: A1. Obtain full-slide images of lung cancer pathology sections; A2. Based on an artificial intelligence model, the whole slide image is analyzed to identify early invasive lesion areas and generate visualization information on the identification basis of the early invasive lesion areas; A3. Receive correction operations performed by multiple pathologists on the early invasive lesion area based on the visualization information of the identification criteria, obtain the correction reasons corresponding to the correction operations, and obtain multiple correction results; A4. Perform a consistency analysis on the corrected results to generate the final early invasive lesion area identification results; A5. Based on the final early invasive lesion area identification results, the artificial intelligence model is incrementally trained; Step A3 includes: A301. Display the identification criteria visualization information to the pathologist as reference information for corrective actions on the early invasive lesion area; A302. Receive the correction operation performed by the pathologist on the early invasive lesion area; the correction operation includes at least one of adding, deleting, adjusting the boundary and modifying the classification of the early invasive lesion area; A303. Based on the type of the correction operation and the image features of the corresponding correction area, extract a list of candidate correction reasons related to the correction operation from a preset pathological feature vocabulary library and display it to the pathologist; A304. Receive the correction reason selected by the pathologist from the candidate correction reason list, and / or receive the supplementary correction reason input by the pathologist; A305. Normalize the selected correction reasons and / or the input supplementary correction reasons to obtain structured correction reasons; A306. Associate the structured correction reasons with the correction operations and record them to obtain the correction results; Step A303 includes: Based on the type of the correction operation and the image features of the corresponding correction area, a preliminary list of candidate correction reasons related to the correction operation is extracted from a preset pathological feature vocabulary. Based on the image features of the correction area, the correlation between each cause in the preliminary extracted list of candidate correction causes and the image features is analyzed to obtain the feature correlation degree of each candidate correction cause. Obtain the co-occurrence frequency information of the correction reasons for early invasive lesion regions with image features similar to the correction region in historical correction data; Combining the feature correlation degree and the co-occurrence frequency information, the preliminary extracted candidate correction reason list is prioritized or grouped to obtain an optimized candidate correction reason list; The optimized list of candidate corrective reasons is displayed to the pathologist.

2. The method for analyzing lung cancer pathological sections based on artificial intelligence according to claim 1, characterized in that, Step A2 includes: A201. Based on the artificial intelligence model, the whole slide image is analyzed to identify the early invasive lesion area and obtain a preliminary identification image containing the boundary of the early invasive lesion area; A202. Extract the response information of the artificial intelligence model to each region in the whole slide image during the process of identifying the early invasive lesion region; A203. Based on the response information, generate an attention map and overlay it on the preliminary identification image to form a map-based identification image, which serves as the visual information for identifying the early invasive lesion area; the color intensity of the attention map indicates the degree of attention the artificial intelligence model pays to each region in the whole slide image.

3. The method for analyzing lung cancer pathological sections based on artificial intelligence according to claim 2, characterized in that, Step A203 includes: Based on the response information, determine the contribution value of each pixel region in the whole slide image to the identification result of the early invasive lesion region; The contribution value is normalized, and the corresponding color is selected from the preset color gradient based on the normalized contribution value to generate the attention map. The attention map is superimposed on the preliminary identification image to form a map-based identification image, which serves as the visual information for identifying the early invasive lesion area.

4. The method for analyzing lung cancer pathological sections based on artificial intelligence according to claim 1, characterized in that, The step of prioritizing or grouping the initially extracted candidate correction reason list by combining the feature correlation degree and the co-occurrence frequency information to obtain an optimized candidate correction reason list includes: Determine the statistical sparsity of the image features in the corrected region; Based on the statistical sparsity, the contribution weights of the feature correlation degree and the co-occurrence frequency information in generating the optimized candidate correction reason list are determined; Based on the contribution weight, combined with the feature correlation degree and the co-occurrence frequency information, the initially extracted candidate correction reason list is prioritized or grouped to obtain the optimized candidate correction reason list.

5. The method for analyzing lung cancer pathological sections based on artificial intelligence according to claim 1, characterized in that, Step A4 includes: A401. Compare the boundary information and classification information of each early invasive lesion region in the multiple correction results, identify the early invasive lesion regions with correction differences, and determine the type of correction difference; A402. Based on the type of correction difference, extract the structured correction reasons corresponding to the early invasive lesion areas with correction differences; A403. Based on the modified difference type and the extracted structured modification reasons, assess the clinical significance of the modified difference; A404. Based on the clinical significance of the correction difference, display to the multiple pathologists the areas of early invasive lesions with correction differences and the corresponding extracted structured correction reasons to assist the pathologists in their discussions; A405. Receive the consensus opinion formed after discussion by the pathologists, or according to the preset fusion rules, combine the clinical importance of the correction differences, and fuse the multiple correction results to generate the final early invasive lesion area identification result information.

6. The method for analyzing lung cancer pathological sections based on artificial intelligence according to claim 5, characterized in that, Step A403 includes: Obtain pathological features of early invasive lesion areas with correction differences, as well as the corresponding patients' clinical background information; Based on the pathological features and clinical background information, the weights of the assessment indicators for the corrected differences in different clinical dimensions are determined from a pre-defined clinical assessment rule base. The clinical significance of the modified difference is calculated based on the modified difference type, the structured modification reason, and the assessment index weight.

7. The method for analyzing lung cancer pathological sections based on artificial intelligence according to claim 6, characterized in that, The step of comprehensively calculating the clinical significance of the corrected difference based on the corrected difference type, the structured correction reason, and the assessment index weight includes: The types of correction differences and the reasons for the structured corrections are converted into corresponding quantitative importance values, respectively. The quantitative importance value is weighted according to the weight of the evaluation index. The clinical importance of the corrected difference is obtained by aggregating and weighting the quantitative importance values.

8. An artificial intelligence-based lung cancer pathology slide analysis system, characterized in that, it identifies early invasive lesions in lung cancer pathology slides based on an artificial intelligence model, and its features include, The system includes: The image acquisition module is used to acquire full-slide images of lung cancer pathology sections; The image analysis module is used to analyze the whole slide image based on an artificial intelligence model, identify the early invasive lesion area, and generate visualization information on the identification basis of the early invasive lesion area. The correction feedback module is used to receive correction operations performed by multiple pathologists on the early invasive lesion area based on the visualization information of the identification criteria, obtain the correction reasons corresponding to the correction operations, and obtain multiple correction results; The consistency analysis module is used to perform consistency analysis on the correction results and generate the final early invasive lesion area identification result information. The model training module is used to incrementally train the artificial intelligence model based on the final early invasive lesion area identification results. When the correction feedback module receives multiple correction operations performed by pathologists on the early invasive lesion area based on the identified visualization information, obtains the correction reasons corresponding to the correction operations, and obtains multiple correction results, it executes the following: A301. Display the identification criteria visualization information to the pathologist as reference information for corrective actions on the early invasive lesion area; A302. Receive the correction operation performed by the pathologist on the early invasive lesion area; the correction operation includes at least one of adding, deleting, adjusting the boundary and modifying the classification of the early invasive lesion area; A303. Based on the type of the correction operation and the image features of the corresponding correction area, extract a list of candidate correction reasons related to the correction operation from a preset pathological feature vocabulary library and display it to the pathologist; A304. Receive the correction reason selected by the pathologist from the candidate correction reason list, and / or receive the supplementary correction reason input by the pathologist; A305. Normalize the selected correction reasons and / or the input supplementary correction reasons to obtain structured correction reasons; A306. Associate the structured correction reasons with the correction operations and record them to obtain the correction results; Step A303 includes: Based on the type of the correction operation and the image features of the corresponding correction area, a preliminary list of candidate correction reasons related to the correction operation is extracted from a preset pathological feature vocabulary. Based on the image features of the correction area, the correlation between each cause in the preliminary extracted list of candidate correction causes and the image features is analyzed to obtain the feature correlation degree of each candidate correction cause. Obtain the co-occurrence frequency information of the correction reasons for early invasive lesion regions with image features similar to the correction region in historical correction data; Combining the feature correlation degree and the co-occurrence frequency information, the preliminary extracted candidate correction reason list is prioritized or grouped to obtain an optimized candidate correction reason list; The optimized list of candidate corrective reasons is displayed to the pathologist.

Images