A brain section pathological report generation system and method
The brain slice pathology report generation system uses a deep learning model to automatically extract and analyze pathology slices, solving the problem of low utilization efficiency of human brain tissue bank resources and achieving efficient and accurate pathology report generation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PEKING UNION MEDICAL COLLEGE HOSPITAL
- Filing Date
- 2025-12-15
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies have low efficiency in utilizing human brain tissue bank resources, relying on manual pathological examinations and data analysis, which are time-consuming, labor-intensive, and easily affected by subjective factors.
Design a brain slice pathology report generation system. Through the collaborative work of the server, the slide reading client, and the management client, the system can automatically extract pathology sub-slices, use a deep learning model to predict pathology results, and combine the primary and predicted pathology results to generate a pathology report.
It improves the efficiency of pathology report generation, reduces the manual marking process, lowers the burden of multi-level review, and ensures the accuracy and consistency of pathology reports.
Smart Images
Figure CN121331344B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical data processing technology, and more specifically, to a system and method for generating brain slice pathology reports. Background Technology
[0002] In current technologies, human brain tissue banks contain a vast number and diverse range of samples, including brain tissue samples from different age groups and disease states. These samples require detailed pathological examinations, molecular biological tests, and other analyses to obtain comprehensive biological information. However, current resource processing relies heavily on manual operations, including sample classification, data entry, and pathological slide interpretation. Specifically, pathologists need to manually interpret a large number of tissue slides, record and analyze pathological features. This process is not only time-consuming and labor-intensive, but the results may also be affected by subjective factors.
[0003] Although existing digital technologies have improved the efficiency of data management to some extent, they still rely heavily on human experience and operation in sample processing and data analysis, which greatly limits the utilization efficiency and sharing effect of human brain tissue bank resources. Summary of the Invention
[0004] The purpose of this application is to provide a brain slice pathology report generation system and method to assist in the brain tissue sample reading process and improve the efficiency of generating pathology reports.
[0005] In a first aspect, the present invention provides a brain slice pathology report generation system, the system comprising at least a server, a slide reading client, and a management client. The management client sends the primary image reading task to the primary image reading client. The primary image reading task includes at least multiple brain tissue pathological sections and section information corresponding to the brain tissue sample. The brain tissue pathology slides in the primary reading task are displayed through the reading interface. The primary reading client performs real-time capture based on the display status of the brain tissue pathology slides in the reading interface to generate at least one first pathology sub-slice. The primary pathology reading client sends the primary pathology results entered by the primary pathologist and the first pathology sub-slice to the server; The server will cut the brain tissue pathology slides to a preset size to generate multiple second pathology sub-slides; The server inputs the first pathological sub-slice and the second pathological sub-slice into the corresponding pathological classification model to obtain the predicted pathological results; The server determines the primary pathological conclusion based on the primary pathological results and predicted pathological results, and then generates a pathological report for the brain tissue sample corresponding to the brain tissue sample by combining the slide information.
[0006] In an optional implementation, the primary slide reading client generates the first pathological sub-slice in the following manner: The real-time magnification of the brain tissue pathology slides in the viewing interface is determined. If the real-time magnification is greater than the preset magnification, the display area of the brain tissue pathology slides in the viewing interface is cropped at a preset frequency to generate at least one first pathology sub-slide; and The display time of the brain tissue pathology slides within the display area is determined. If the display time exceeds the preset time, the display area of the brain tissue pathology slides in the slide viewing interface is cropped to generate at least one first pathology sub-slide.
[0007] In an optional implementation, the server obtains the predicted pathology results in the following ways: The first pathological sub-slice with a magnification greater than the preset magnification is input into the first feature extraction network to obtain the corresponding high-magnification feature vector; The first and second pathological sub-slices with magnification less than the preset magnification are input into the second feature extraction network to obtain the corresponding low-magnification feature vectors. All high-magnification and low-magnification feature vectors are input into a cross-scale feature fusion network to obtain fused features; The fused features are input into the pathological enhancement network to obtain enhanced features; The enhanced features are input into the classifier to obtain predicted pathological outcomes.
[0008] In an optional implementation, the first feature extraction network includes multiple residual layers and spatial pyramid pooling layers, and the second feature extraction network includes multiple residual layers, downsampling layers, and spatial pyramid pooling layers. Cross-scale feature fusion networks include cross-attention fusion layers, weighted fusion layers, and linear layers.
[0009] In an optional implementation, the primary image reading client obtains the primary image reading conclusion of the brain tissue pathology slides input by the primary image reader and sends it to the server; If the initial review conclusions of all brain tissue pathology sections corresponding to the brain tissue sample are all "review passed", then the server will execute the step of generating a brain tissue sample pathology report. For brain tissue pathology slides whose initial review conclusion is "not approved", the server generates a corresponding secondary review task and sends it to the secondary review client to obtain the secondary review conclusion input by the secondary reviewer and the secondary pathology conclusion generated based on the initial pathology conclusion. If all secondary slide reviews of the brain tissue samples corresponding to the brain tissue sample are deemed satisfactory, the server will then proceed to generate a brain tissue sample pathology report.
[0010] In an optional implementation, the server determines whether the secondary image reading client has a specific brain bank identifier; If so, the server will determine whether a third-level image review is required based on the secondary pathology conclusion; If necessary, the server generates a corresponding Level 3 image reading task and sends it to the Level 3 image reading client to obtain the Level 3 image reading conclusion and Level 3 pathology conclusion entered by the Level 3 image reader.
[0011] In an optional implementation, if the level 3 pathological conclusion of the brain tissue sample is passed, the server sends the corresponding brain tissue sample pathology report to the management client. The management client stores the pathology report of the brain tissue sample corresponding to the brain tissue sample in the brain tissue database.
[0012] In an optional implementation, the server generates a pathological report for the brain tissue sample based on the primary / secondary / tertiary pathological conclusions, slide reading process information, slide information corresponding to the brain tissue sample, and a preset table template.
[0013] Secondly, the present invention provides a method for generating a brain slice pathology report, the method comprising: The management client sends the primary image reading task to the primary image reading client. The primary image reading task includes at least multiple brain tissue pathological sections and section information corresponding to the brain tissue sample. The brain tissue pathology slides in the primary reading task are displayed through the reading interface. The primary reading client performs real-time capture based on the display status of the brain tissue pathology slides in the reading interface to generate at least one first pathology sub-slice. The primary pathology reading client sends the primary pathology results entered by the primary pathologist and the first pathology sub-slice to the server; The server will cut the brain tissue pathology slides to a preset size to generate multiple second pathology sub-slides; The server inputs the first pathological sub-slice and the second pathological sub-slice into the corresponding pathological classification model to obtain the predicted pathological results; The server determines the primary pathological conclusion based on the primary pathological results and predicted pathological results, and then generates a pathological report for the brain tissue sample corresponding to the brain tissue sample by combining the slide information.
[0014] In an optional implementation, the server obtains the predicted pathology results in the following ways: The first pathological sub-slice with a magnification greater than the preset magnification is input into the first feature extraction network to obtain the corresponding high-magnification feature vector; The first and second pathological sub-slices with magnification less than the preset magnification are input into the second feature extraction network to obtain the corresponding low-magnification feature vectors. All high-magnification and low-magnification feature vectors are input into a cross-scale feature fusion network to obtain fused features; The fused features are input into the pathological enhancement network to obtain enhanced features; The enhanced features are input into the classifier to obtain predicted pathological outcomes.
[0015] This application provides a brain tissue pathology report generation system and method. The system includes at least a server, a reading client, and a management client. The management client sends a primary reading task to the primary reading client. The primary reading task includes at least multiple brain tissue pathology sections and section information corresponding to the brain tissue sample. The brain tissue pathology sections from the primary reading task are displayed through a reading interface. The primary reading client performs real-time cropping based on the display status of the brain tissue pathology sections in the reading interface to generate at least one first pathology sub-section. The primary reading client sends the primary pathology results input by the primary reader and the first pathology sub-section to the server. The server crops the brain tissue pathology sections according to a preset size to generate multiple second pathology sub-sections. The server inputs the first and second pathology sub-sections into the corresponding pathology classification model to obtain predicted pathology results. Based on the primary pathology results and predicted pathology results, the server determines the primary pathology conclusion and generates a brain tissue sample pathology report corresponding to the brain tissue sample by combining the section information. Compared with existing technologies, the technical solution of this application, based on a graded slide reading mechanism, rationally allocates slide reading tasks. In the initial slide reading stage, it detects the slide browsing status of the primary slide reader, identifies the slide parts that the primary slide reader focuses on, and inputs them into the pathological classification model to obtain predicted pathological results. These predicted pathological results are then fused with the judgment results of the primary slide reader to obtain the predicted pathological results. This not only eliminates the marking process for primary slide readers but also allows for the verification and guidance of the initial pathological results based on the predicted pathological results, reducing the burden of reviewing slides at other levels and improving slide reading efficiency. Attached Figure Description
[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a schematic diagram of the brain slice pathology report generation system provided in an embodiment of this application; Figure 2 The communication flowchart of the brain slice pathology report generation system provided in the embodiments of this application is shown. Detailed Implementation
[0018] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.
[0019] Example 1 In one embodiment of this application, a brain slice pathology report generation system is provided. The system includes a server, a slide reading client, and a management client.
[0020] The server and management client can be deployed on one side of the human brain tissue bank's digital management platform. Brain bank administrators can use the management client to assign image reading tasks for brain tissue samples uploaded to the system. The server is used to provide intelligent data processing and resource integration.
[0021] The image reading clients can include primary, secondary, tertiary, and quaternary clients. The primary client can be deployed on the side of brain bank staff or doctors for initial image reading of brain tissue samples.
[0022] The secondary image reading client can be deployed on the pathology expert's side to perform secondary image reading of brain tissue samples, while also reviewing the results of the primary image reading.
[0023] The Level 3 and Level 4 image reading clients can be deployed on either the pathology expert's side or the quality control expert's side for further review of tasks requiring Level 3 image reading, as well as for random checks and quality control of past image reading tasks.
[0024] Primary level radiologists can annotate images, fill in scores, and provide preliminary pathological results for brain tissue pathology slides corresponding to primary radiology tasks. Secondary level radiologists, after reviewing all brain tissue pathology slides corresponding to the brain tissue sample and verifying the primary radiology results, can determine whether to generate a pathology report. In special cases, they can apply for tertiary radiology review.
[0025] The brain slice pathology report generation system provided in this application constructs a complete graded reading mechanism compared with the prior art, which can ensure the quality of brain bank pathology samples.
[0026] Example 2 Because there are a large number of brain tissue samples, and each brain tissue sample corresponds to multiple brain tissue pathological sections, this application provides an intelligent image reading method in one embodiment to reduce the burden of image reading and improve efficiency. The method specifically includes: S1. The management client sends the primary image reading task to the primary image reading client. The primary image reading task includes at least multiple brain tissue pathological sections and section information corresponding to the brain tissue sample.
[0027] Brain bank administrators can first distribute the film reading tasks in the system to the primary film reading clients, so that primary film readers can read the films.
[0028] Here, multiple brain tissue pathological sections corresponding to a single brain tissue sample can form one or more primary slide reading tasks. Multiple primary slide reading tasks can be assigned to the same primary slide reader or to different primary slide readers.
[0029] The initial slide reading task here can include one or more brain tissue pathological slides and slide information. Slide information can include sample number, sampling number, gender, sampling site, sampler, sampling time, condition, allocation time, staining method, and sectioning method.
[0030] S2. The brain tissue pathology slides in the primary slide reading task are displayed through the slide reading interface. The primary slide reading client performs real-time capture based on the display status of the brain tissue pathology slides in the slide reading interface to generate at least one first pathology sub-slide.
[0031] The image viewing client can provide an interface for image viewing, allowing users to view brain tissue pathology slides and perform operations such as zooming and moving the slides.
[0032] During the slide reading process, the basic slide reading client can generate the first pathology sub-slice in the following ways: The real-time magnification of the brain tissue pathology slides in the viewing interface is determined. If the real-time magnification is greater than the preset magnification, the display area of the brain tissue pathology slides in the viewing interface is cropped at a preset frequency to generate at least one first pathology sub-slide; and The display time of the brain tissue pathology slides within the display area is determined. If the display time exceeds the preset time, the display area of the brain tissue pathology slides in the slide viewing interface is cropped to generate at least one first pathology sub-slide.
[0033] Existing image reading software typically provides a marking function for viewers to mark special areas. However, this not only requires manual operation but is also prone to omissions due to the viewer's subjective judgment, thus affecting the accuracy of the image reading.
[0034] Therefore, in step S2, the slide portion observed by the reader for an extended period and the slide portion viewed under magnification are automatically captured by the software to form at least one first pathological slide. For example, the preset magnification can be 20x and the preset duration can be 30 seconds.
[0035] This allows for a more comprehensive capture of the key areas of interest or areas of concern to the spectators. Based on these pathological sub-slices, identification can be performed using a deep learning model, resulting in more accurate classification results.
[0036] S3. The primary slide reading client sends the primary pathology results entered by the primary slide reader and the first pathology sub-slice to the server.
[0037] The primary pathology findings here are preliminary assessments and descriptive records of the pathology, such as, "Alzheimer's disease, with numerous Aβ plaques."
[0038] S4. The server will cut the brain tissue pathological slices to a preset size to generate multiple second pathological sub-slices.
[0039] In step S4, the brain tissue pathological sections are cut. This can be done through random sampling or by using target recognition techniques to identify and cut the lesion areas; there are no limitations. The second pathological sub-sections may partially overlap or not overlap at all.
[0040] The second pathological subsection here is the same size as the first pathological subsection.
[0041] S5. The server inputs the first pathological sub-slice and the second pathological sub-slice into the corresponding pathological classification model to obtain the predicted pathological results.
[0042] Specifically, the first pathological sub-slice with a magnification greater than a preset magnification is input into the first feature extraction network to obtain the corresponding high-magnification feature vector; the first and second pathological sub-slices with a magnification less than the preset magnification are input into the second feature extraction network to obtain the corresponding low-magnification feature vector; all high-magnification and low-magnification feature vectors are input into the cross-scale feature fusion network to obtain fused features; the fused features are input into the pathology enhancement network to obtain enhanced features; and the enhanced features are input into the classifier to obtain the predicted pathology result.
[0043] The pathological classification model here may include a first feature extraction network, a second feature extraction network, a cross-scale feature fusion network, a pathological enhancement network, and a classifier. The first feature extraction network includes multiple residual layers and spatial pyramid pooling layers, while the second feature extraction network includes multiple residual layers, downsampling layers, and spatial pyramid pooling layers.
[0044] The first and second feature extraction networks are based on fully convolutional networks such as ResNet-50 and ConvNeXt. Residual layers alleviate the vanishing gradient problem in deep networks through skip connections. For pathological sub-slices with low magnification, additional downsampling layers, such as convolutions or pooling with stride=2, are added to further extract detailed features. Simultaneously, to integrate local features from pathological sub-slices of different sizes or locations, Spatial Pyramid Pooling (SPP) layers (e.g., 4×4, 8×8, 16×16) are employed to unify multi-scale features, outputting a fixed-dimensional feature vector and integrating local features from different regions.
[0045] Cross-scale feature fusion networks include cross-attention fusion layers, weighted fusion layers, and linear layers.
[0046] In the feature fusion stage, a cross-attention mechanism is adopted, with the high-multiplied feature vector as Query (Q) and the low-multiplied feature vector as Key (K) and Value (V). The attention weights Attention(Q,K,V) are calculated as softmax(Q,K,V). It dynamically selects low-magnification areas that require attention under high-magnification features. This simulates the pathologist's reading logic of "locating lesions under high magnification and observing the overall picture under low magnification."
[0047] The weighted fusion layer can perform a weighted summation of the attention-weighted features and the original high / low magnification features: F=α· + +γ· ; Where α, β, γ are learnable parameters. It is a high-multiplier feature vector. These are low-magnification feature vectors. This is the output vector of the attention mechanism.
[0048] Linear layers are used to map the fused features to a unified semantic space, such as fully connected layers that reduce the dimensionality to 512 dimensions.
[0049] Pathological enhancement networks can employ channel attention mechanisms, such as the SE module, to recalibrate feature channel weights and output enhanced features (with dimensions consistent with the fused features).
[0050] The classifier structure can include fully connected layers and a softmax layer. It outputs the probability distribution for each class.
[0051] By capturing high-magnification feature vectors that are of primary interest to the radiologist, and interacting with low-magnification feature vectors from multiple modalities, the system dynamically selects the key areas of interest, thereby obtaining fused features. These fused features are then enhanced to form the final enhanced features used for classification, further improving the accuracy of pathological classification predictions.
[0052] S6. The server determines the primary pathological conclusion based on the primary pathological results and predicted pathological results, and generates a pathological report for the brain tissue sample corresponding to the brain tissue sample by combining the slide information.
[0053] In step S6, for each brain tissue pathology slide, the server can determine whether the preliminary pathology result and the predicted pathology result match. If they do not match, the preliminary slide review is deemed unsuccessful and requires further review. The predicted pathology result here refers to the disease diagnosis predicted by the model based on the brain tissue pathology sub-slide, such as Alzheimer's disease or Parkinson's disease. The server can compare the disease diagnosis in the preliminary pathology result with the predicted pathology result to see if they are the same, and determine whether the descriptive text in the preliminary pathology result matches the predicted pathology result. If both are correct, the preliminary pathology result can be used as the preliminary pathology conclusion for subsequent brain tissue sample pathology reports.
[0054] It should be noted that the slide review process also requires verification of other standards to ensure the quality and accuracy of the samples entering the database. This includes checking whether sample numbers match, whether staining quality is acceptable, whether slide shapes are symmetrical and complete, and whether pathological slide images are clear. If all review criteria are met, the initial slide review can be considered passed.
[0055] If all brain tissue pathology slides corresponding to a brain tissue sample pass the initial slide review, a brain tissue sample pathology report can be generated.
[0056] This application provides a brain slide pathology report generation system. Based on a hierarchical slide reading mechanism, the system rationally allocates slide reading tasks. In the initial slide reading stage, it detects the slide browsing status of novice slide readers, identifies the slide sections that novice slide readers focus on, and inputs them into a pathology classification model to obtain predicted pathology results. These predicted pathology results are then fused with the novice slide readers' judgments to obtain the predicted pathology results. This not only eliminates the need for novice slide readers to mark slides but also allows for the verification and guidance of the initial pathology results based on the predicted pathology results.
[0057] Compared to existing technologies that require multiple levels of review for slide reading, the technical solution of this application eliminates a significant amount of secondary slide reading procedures by using decision-making based on primary pathological results and predicted pathological results. It also quickly identifies brain tissue pathological slides requiring secondary review, reducing the burden of reviewing other levels and improving efficiency. Furthermore, it provides a reference for subsequent slide readings when pathological results are inconsistent.
[0058] In one feasible embodiment, the primary image reader client obtains the primary image reader's preliminary image reading conclusion for the brain tissue pathology slides and sends it to the server. If the primary image reading conclusions for all brain tissue pathology slides corresponding to the brain tissue sample are all deemed satisfactory, the server executes the step of generating a brain tissue sample pathology report.
[0059] If all brain tissue pathology slides of the same brain tissue sample have passed the initial slide review, a brain tissue sample pathology report can be automatically generated by the server.
[0060] For brain tissue pathology slides whose initial review conclusion is "not acceptable," the server generates a corresponding secondary review task and sends it to the secondary review client to obtain the secondary review conclusion input by the secondary reviewer and the secondary pathology conclusion generated based on the initial pathology conclusion.
[0061] If there are doubts about the pathology results or problems with the slide quality, the slides are sent to the secondary reading client for review. At the secondary reading node, slides with quality issues can be returned. Secondary readers can also re-evaluate the pathology results, referencing both the primary and predicted pathology results to arrive at a secondary reading conclusion.
[0062] If all secondary slide reviews of the brain tissue samples corresponding to the brain tissue sample are deemed satisfactory, the server will then proceed to generate a brain tissue sample pathology report.
[0063] This ensures the standardization of brain tissue samples, ultimately leading to the formation of a brain tissue sample pathology report, which allows the sample information to be stored in a database for subsequent research or other uses.
[0064] Furthermore, in one feasible implementation, the secondary image reading client is deployed on the pathology expert's side. However, due to limited resources and differences in expert skill levels, a brain bank identifier is added to the certified parts of the pathology expert's work. If the secondary image reading is performed by a pathology expert with a brain bank identifier, further review is unnecessary. Specifically, the server can determine whether the secondary image reading client has a brain bank identifier. If so, the server determines whether a tertiary image reading review is required based on the secondary pathology conclusion. If required, the server generates a corresponding tertiary image reading task and sends it to the tertiary image reading client to obtain the tertiary image reading conclusion and tertiary pathology conclusion entered by the tertiary image reader.
[0065] If the level 3 pathological conclusion of the brain tissue sample is satisfactory, the server sends the corresponding brain tissue sample pathology report to the management client. The management client then stores the corresponding brain tissue sample pathology report in the brain tissue database.
[0066] It should be noted that the Level 3 review here can also review samples that the Level 2 review deems require further review.
[0067] In one feasible implementation, the server can generate a pathological report for the brain tissue sample based on the primary / secondary / tertiary pathological conclusions, slide reading process information, slide information corresponding to the brain tissue sample, and a preset table template.
[0068] The pathology report for brain tissue samples can be in tabular form. The server can automatically generate the report based on the collected data. The report may include the sample number, age, gender, time of death, sampling delay, cerebrospinal fluid pH, cause of death, clinical diagnosis, family history, toxicology history, the person who generated the report, the report generation time, the person who reviewed the slides, the review time, the pathology results (description), and the review conclusion.
[0069] Example 3 In one embodiment of this application, a method for generating a brain slice pathology report is provided, the method comprising: The management client sends the primary image reading task to the primary image reading client. The primary image reading task includes at least multiple brain tissue pathological sections and section information corresponding to the brain tissue sample. The brain tissue pathology slides in the primary reading task are displayed through the reading interface. The primary reading client performs real-time capture based on the display status of the brain tissue pathology slides in the reading interface to generate at least one first pathology sub-slice. The primary pathology reading client sends the primary pathology results entered by the primary pathologist and the first pathology sub-slice to the server; The server will cut the brain tissue pathology slides to a preset size to generate multiple second pathology sub-slides; The server inputs the first pathological sub-slice and the second pathological sub-slice into the corresponding pathological classification model to obtain the predicted pathological results; The server determines the primary pathological conclusion based on the primary pathological results and predicted pathological results, and then generates a pathological report for the brain tissue sample corresponding to the brain tissue sample by combining the slide information.
[0070] In an optional implementation, the server obtains the predicted pathology results in the following ways: The first pathological sub-slice with a magnification greater than the preset magnification is input into the first feature extraction network to obtain the corresponding high-magnification feature vector; The first and second pathological sub-slices with magnification less than the preset magnification are input into the second feature extraction network to obtain the corresponding low-magnification feature vectors. All high-magnification and low-magnification feature vectors are input into a cross-scale feature fusion network to obtain fused features; The fused features are input into the pathological enhancement network to obtain enhanced features; The enhanced features are input into the classifier to obtain predicted pathological outcomes.
[0071] The method for generating a brain slice pathology report provided in this application can be referred to in terms of principle and technical effect as the brain slice pathology report generation system in the foregoing embodiments, and will not be repeated here.
[0072] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.
[0073] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0074] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0075] It should be noted that if the function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0076] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.
[0077] 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. A brain slice pathology report generation system, characterized in that, The system includes at least a server, a viewing client, and a management client. The management client sends the primary image reading task to the primary image reading client. The primary image reading task includes at least multiple brain tissue pathological sections and section information corresponding to the brain tissue sample. The primary image reading client determines the real-time magnification of the brain tissue pathological slices in the image reading interface based on the image reading operation of the image reader. If the real-time magnification is greater than the preset magnification, the display area of the brain tissue pathological slices in the image reading interface is cropped according to the preset frequency to generate at least one first pathological sub-slice. The primary image reading client also determines the dwell time of the brain tissue pathological slices in the display area based on the image reading operation of the image reader. If the dwell time is greater than the preset time, the display area of the brain tissue pathological slices in the image reading interface is cropped to generate at least one first pathological sub-slice. The primary pathology reading client sends the primary pathology results entered by the primary pathologist and the first pathology sub-slice to the server; The server will cut the brain tissue pathology slides to a preset size to generate multiple second pathology sub-slides; The server inputs the first pathological sub-slice and the second pathological sub-slice into the corresponding pathological classification model to obtain the predicted pathological results; The server determines the primary pathological conclusion based on the preliminary pathological results and predicted pathological results, and then generates a corresponding brain tissue sample pathology report by combining the slide information. The server obtains the predicted pathology results through the following methods: The first pathological sub-slice with a magnification greater than the preset magnification is input into the first feature extraction network to obtain the corresponding high-magnification feature vector; The first and second pathological sub-slices with magnification less than the preset magnification are input into the second feature extraction network to obtain the corresponding low-magnification feature vectors. All high-magnification and low-magnification feature vectors are input into a cross-scale feature fusion network to obtain fused features; The fused features are input into the pathological enhancement network to obtain enhanced features; The enhanced features are input into the classifier to obtain predicted pathological outcomes.
2. The system according to claim 1, characterized in that, The first feature extraction network includes multiple residual layers and spatial pyramid pooling layers, and the second feature extraction network includes multiple residual layers, downsampling layers, and spatial pyramid pooling layers. Cross-scale feature fusion networks include cross-attention fusion layers, weighted fusion layers, and linear layers.
3. The system according to claim 1, characterized in that, The primary image reading client obtains the primary image reading conclusions of the brain tissue pathology slides input by the primary image reader and sends them to the server; If the initial review conclusions of all brain tissue pathology sections corresponding to the brain tissue sample are all "review passed", then the server will execute the step of generating a brain tissue sample pathology report. For brain tissue pathology slides whose initial review conclusion is "not approved", the server generates a corresponding secondary review task and sends it to the secondary review client to obtain the secondary review conclusion input by the secondary reviewer and the secondary pathology conclusion generated based on the initial pathology conclusion. If all secondary slide reviews of the brain tissue samples corresponding to the brain tissue sample are deemed satisfactory, the server will then proceed to generate a brain tissue sample pathology report.
4. The system according to claim 3, characterized in that, The server determines whether the secondary image reading client has a specific brain database identifier; If so, the server will determine whether a third-level image review is required based on the secondary pathology conclusion; If necessary, the server generates a corresponding Level 3 image reading task and sends it to the Level 3 image reading client to obtain the Level 3 image reading conclusion and Level 3 pathology conclusion entered by the Level 3 image reader.
5. The system according to claim 4, characterized in that, If the level 3 pathological conclusion of the brain tissue sample is approved, the server will send the corresponding brain tissue sample pathology report to the management client. The management client stores the pathology report of the brain tissue sample corresponding to the brain tissue sample in the brain tissue database.
6. The system according to claim 5, characterized in that, The server generates a pathology report for the brain tissue sample based on the primary / secondary / tertiary pathology conclusions, slide reading process information, slide information corresponding to the brain tissue sample, and preset table templates.
7. A method for generating a brain slice pathology report, characterized in that, The method of the brain slice pathology report generation system according to any one of claims 1-6 includes: The management client sends the primary image reading task to the primary image reading client. The primary image reading task includes at least multiple brain tissue pathological sections and section information corresponding to the brain tissue sample. The brain tissue pathology slides in the primary reading task are displayed through the reading interface. The primary reading client performs real-time capture based on the display status of the brain tissue pathology slides in the reading interface to generate at least one first pathology sub-slice. The primary pathology reading client sends the primary pathology results entered by the primary pathologist and the first pathology sub-slice to the server; The server will cut the brain tissue pathology slides to a preset size to generate multiple second pathology sub-slides; The server inputs the first pathological sub-slice and the second pathological sub-slice into the corresponding pathological classification model to obtain the predicted pathological results; The server determines the primary pathological conclusion based on the primary pathological results and predicted pathological results, and then generates a pathological report for the brain tissue sample corresponding to the brain tissue sample by combining the slide information.
8. The method according to claim 7, characterized in that, The server obtains the predicted pathology results in the following ways: The first pathological sub-slice with a magnification greater than the preset magnification is input into the first feature extraction network to obtain the corresponding high-magnification feature vector; The first and second pathological sub-slices with magnification less than the preset magnification are input into the second feature extraction network to obtain the corresponding low-magnification feature vectors. All high-magnification and low-magnification feature vectors are input into a cross-scale feature fusion network to obtain fused features; The fused features are input into the pathological enhancement network to obtain enhanced features; The enhanced features are input into the classifier to obtain predicted pathological outcomes.