Pathological image review method and device based on structural counterfactual raw observation separability

Abstract

The application provides a pathological image auditing method and device based on structural counterfactual original observation separability, which comprises the following steps: obtaining a scanning original frame and scanning metadata of a pathological scanner, and performing multiple candidate reconstructions after processing by using a generative recovery model; constructing a device degradation parameter based on the scanning metadata, simulating a simulated observation image for the candidate reconstructed image, calculating a robust observation distance according to the image error between the simulated observation image and the scanning original frame, and screening an observation equivalent recovery solution; setting positive and negative structure constraints for each pathological structure, and obtaining an optimal positive recovery solution and an optimal negative recovery solution that satisfy the constraints and have the minimum robust observation distance; and constructing an evidence margin between the optimal positive recovery solution and the optimal negative recovery solution, and outputting an auditing result of the pathological structure according to the evidence margin. In the scheme, the pathological structure recovery is converted into two competing hypotheses, the evidence margin is constructed through the robust observation distance under multiple device degradation parameters, and it is determined whether the structure is truly supported by the original observation, so that the generative illusion can be accurately identified and the optimal supplementary sampling can be guided.

Description

Technical Field

[0001] This application relates to the field of pathological image processing, and in particular to a method and apparatus for reviewing pathological images based on the separability of original observations of structural counterfactuality. Background Technology

[0002] Digital scanning of pathological slides typically utilizes microscope objectives, illumination sources, image sensors, XY motion platforms, Z-axis focusing mechanisms, and image stitching systems to continuously image tissue regions on a glass slide and create a whole-slide digital image. However, due to factors such as objective defocus, platform movement, mechanical vibration, underexposure, uneven illumination, sensor noise, field aberrations, image stitching errors, and data compression, the actual pathological images obtained from scanning often suffer from various quality degradations, including defocus, motion blur, noise, streaks, chromatic aberration, insufficient resolution, or stitching seams.

[0003] To improve image quality, existing technologies have begun to employ generative methods such as convolutional neural networks, generative adversarial networks, diffusion models, conditional generative models, and visual fundamental models to perform deblurring, denoising, super-resolution reconstruction, digital refocusing, illumination correction, color restoration, or joint repair of multiple artifacts on pathological images. Generative restoration models can use image priors learned from training data to supplement missing high-frequency details, thereby obtaining visually clearer restored images. However, precisely because these models rely on prior data to generate content, the details in their output are not necessarily uniquely determined by the current original sensor observations, and the restored results may introduce structural information lacking support from the original data.

[0004] For pathological images with significant information loss, the same low-resolution, low signal-to-noise ratio, or blurry original image often corresponds to multiple clear images from different but similarly sensor-observed sources. In such cases, generative reconstruction models may output visually natural but lack structural content supported by the original observations; this phenomenon is commonly referred to as generative illusion. Generative illusion directly affects the accuracy of lesion grading, cell counting, and morphological interpretation, potentially leading to misdiagnosis or missed diagnosis.

[0005] Existing methods for assessing the quality of restored images typically employ metrics such as peak signal-to-noise ratio (PSNR), structural similarity, perceptual distance, human visual scoring, or pathological feature model scoring. While these metrics can evaluate whether the output image is clear or close to the reference image, they are insufficient to prove whether a specific cell nucleus, gland, or membrane structure in the restored image is necessary to interpret the original sensor data. Some existing technologies further incorporate data consistency constraints during the training or inference process of the restoration model, reprojecting the restoration results back to the input domain through a pre-defined degradation model and requiring them to approximate the original image. Other technologies evaluate the stability of virtual stained images through forward and backward generation loops, or compare cell instances and tissue topology before and after restoration. However, these schemes still have significant shortcomings: First, they typically only review one deterministic restored image output by the generative model, failing to consider that the same original observation may correspond to multiple equivalent restored solutions, thus failing to distinguish between "the structure is uniquely supported by the original observation" and "the structure is randomly selected only by the generative prior"; second, they usually use a fixed set of optical point spread functions, motion kernels, and noise parameters, while actual scanning parameters have calibration errors, temporal drift, and spatial variations, and erroneous results may pass the review by using a set of degenerate parameters that are favorable to them; third, they directly extract cell nuclei from the blurred original image and instantiate them with the restored image. During matching, the detection error of the original image itself may be misjudged as the addition, deletion or split of the restored image; fourth, when using multiple quality scores for linear weighting or full-image averaging, a small number of high-risk structures may be diluted by a large number of normal background areas; fifth, the existing scheme only judges whether the restoration result passes the review, without predicting how much structural ambiguity can be eliminated for different supplementary acquisition actions, and fixed rules such as refocusing when out of focus and decelerating when motion blur is not necessarily the lowest cost verification method for the current suspicious structures; sixth, there is a lack of a complete chain of evidence from the original frame, observation equivalent solution, counterfactual evidence of structure, supplementary acquisition actions to the final display state.

[0006] Therefore, there is an urgent need for a method that can perform structured review of generatively restored pathological images, identify high-risk pathological structures that lack support from the original observations, and guide optimal supplementary acquisition decisions, thereby addressing the risk of diagnostic errors caused by hallucinations in existing generative pathological image restoration. Summary of the Invention

[0007] This application provides a method and apparatus for reviewing pathological images based on the separability of original observations of counterfactual structures. This approach transforms pathological structure restoration into two competing hypotheses, constructs an evidence margin through robust observation distances under multiple sets of equipment degradation parameters, and determines whether the structure is truly supported by the original observations, thereby accurately identifying generative illusions and guiding optimal re-sampling.

[0008] In a first aspect, embodiments of this application provide a method for reviewing pathological images based on the separability of original observations of structural counterfactuality, the method comprising: The original scan frame and corresponding scan metadata of the pathology scanner are obtained. The original scan frame is processed using a generative recovery model to obtain the image to be recovered for review. The image to be recovered for review is then reconstructed multiple times to obtain multiple candidate reconstructed images. Multiple sets of device degradation parameters are constructed based on scan metadata. Each candidate reconstructed image is simulated using each set of device degradation parameters to generate multiple simulated observation images. The robust observation distance of each candidate reconstructed image is calculated. Candidate reconstructed images with robust observation distance less than or equal to the observation equivalence determination threshold are taken as observation equivalence recovery solutions. For each candidate reconstructed image, the robust observation distance of the corresponding candidate reconstructed image is obtained based on the image error between each corresponding simulated observation image and the original scan frame. For each pathological structure in the image to be reviewed and restored, positive and negative structural constraints are set. The positive structural constraints are used to constrain the corresponding pathological structure to maintain its current structural state, and the negative structural constraints are used to constrain the corresponding pathological structure to have structural modifications. Obtain the optimal positive recovery solution and the optimal negative recovery solution for each pathological structure. The optimal positive recovery solution is the candidate reconstructed image that satisfies the corresponding positive structure constraint and has the smallest robust observation distance. The optimal negative recovery solution is the candidate reconstructed image that satisfies the corresponding negative structure constraint and has the smallest robust observation distance. For any pathological structure, an evidence margin is constructed between the corresponding optimal positive recovery solution and the optimal negative recovery solution. The evidence margin represents the difference in robust observation distance between the corresponding optimal negative recovery solution and the optimal positive recovery solution. Based on the evidence margin, the review result of the corresponding pathological structure in the candidate reconstructed image is obtained.

[0009] Secondly, embodiments of this application provide a pathological image review device based on the separability of original structural counterfactual observations, comprising: The acquisition module is used to acquire the original scan frame of the pathology scanner and the corresponding scan metadata, process the original scan frame using a generative recovery model to obtain the image to be recovered for review, and perform multiple candidate reconstructions on the image to be recovered to obtain multiple candidate reconstructed images. The simulation module constructs multiple sets of device degradation parameters based on scan metadata. Each candidate reconstructed image is simulated using each set of device degradation parameters to generate multiple simulated observation images. The robust observation distance of each candidate reconstructed image is calculated. Candidate reconstructed images with robust observation distances less than or equal to the observation equivalence determination threshold are taken as observation equivalence recovery solutions. For each candidate reconstructed image, the robust observation distance of the corresponding candidate reconstructed image is obtained based on the image error between each corresponding simulated observation image and the original scan frame. The constraint construction module sets positive and negative structural constraints for each pathological structure in the image to be reviewed and restored. The positive structural constraints are used to constrain the corresponding pathological structure to maintain its current structural state, and the negative structural constraints are used to constrain the corresponding pathological structure from having structural modifications. The constraint module obtains the optimal positive recovery solution and the optimal negative recovery solution for each pathological structure. The optimal positive recovery solution is the candidate reconstructed image that satisfies the corresponding positive structure constraint and has the smallest robust observation distance. The optimal negative recovery solution is the candidate reconstructed image that satisfies the corresponding negative structure constraint and has the smallest robust observation distance. The review module constructs an evidence margin between the corresponding optimal positive recovery solution and the optimal negative recovery solution for any pathological structure. The evidence margin represents the difference in robust observation distance between the corresponding optimal negative recovery solution and the optimal positive recovery solution. Based on the evidence margin, the review result of the corresponding pathological structure in the candidate reconstructed image is obtained.

[0010] Thirdly, embodiments of this application provide an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform a pathological image review method based on the structural counterfactual original observation separability.

[0011] The main contributions and innovations of this invention are as follows: This application embodiment randomly samples multiple sets of device degradation parameters based on the calibration benchmark values ​​and allowable variation ranges of each dimension of the scan metadata. The device degradation operator projects candidate reconstructed images back to the sensor domain, generating multiple sets of simulated observation images. This incorporates device calibration errors, temporal drift, and spatial variations into the review process, avoiding the loophole in existing technologies that use only one set of fixed degradation parameters, allowing erroneous results to pass review by utilizing favorable parameters. This application embodiment uses the noise covariance corresponding to each set of degradation parameters to whiten and normalize the error between simulated observations and the original frame. Then, robust summarization operators such as maximum value / high quantile / conditional risk value are used to integrate and obtain robust observation distances, thereby eliminating the influence of noise level differences under different degradation parameters. This ensures that images with "small distances only under a few favorable parameters and large distances under other reasonable parameters" still have relatively large robust distances, preventing them from passing review by relying on a single set of favorable parameters and objectively assessing the physical feasibility of image restoration. This application embodiment separately... By setting positive and negative structure constraints, the optimal positive and negative recovery solutions with the minimum robust observation distance are obtained in their respective local neighborhoods. This transforms the structure verification into two competing hypotheses: "preserving the structure" and "changing the structure." The constraints only apply to the local neighborhood of the target structure, while remaining consistent in other regions. This ensures that the subsequent margin accurately reflects the uncertainty of the target structure itself, eliminates the interference of structural changes in irrelevant regions on the results, and avoids misjudgment of the original image's own detection error. In this embodiment, the difference between the robust observation distances of the optimal negative and optimal positive recovery solutions is constructed as an evidentiary margin. This margin is compared with a credibility threshold to output a conclusion that the structure is credible or uncredible. A sufficiently large margin indicates that maintaining the original structure can significantly better explain the original frame, thus determining credibility. An insufficient margin or a negative margin indicates that the original observation cannot distinguish between the two states of preservation and modification, thus determining uncredibility. This achieves accurate identification of generative illusions, and the judgment is made item by item according to the structure rather than averaging across the entire image, avoiding the dilution of high-risk structures by a large amount of normal background.

[0012] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a pathological image review method based on the separability of original observations of structural counterfactuality, according to an embodiment of this application; Figure 2 This is a structural block diagram of a pathological image review device based on the separability of original observations of structural counterfactuality, according to an embodiment of this application. Figure 3This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0015] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0016] Example 1 This application provides a method for reviewing pathological images based on the separability of original counterfactual observations. It deconstructs pathological structures into two competing hypotheses, constructs an evidence margin using robust observation distances under multiple sets of device degradation parameters, and determines whether the structure is truly supported by the original observations. This accurately identifies generative hallucinations and guides optimal re-sampling. Specifically, refer to... Figure 1 The method includes: The original scan frame and corresponding scan metadata of the pathology scanner are obtained. The original scan frame is processed using a generative recovery model to obtain the image to be recovered for review. The image to be recovered for review is then reconstructed multiple times to obtain multiple candidate reconstructed images. Multiple sets of device degradation parameters are constructed based on scan metadata. Each candidate reconstructed image is simulated using each set of device degradation parameters to generate multiple simulated observation images. The robust observation distance of each candidate reconstructed image is calculated. Candidate reconstructed images with robust observation distance less than or equal to the observation equivalence determination threshold are taken as observation equivalence recovery solutions. For each candidate reconstructed image, the robust observation distance of the corresponding candidate reconstructed image is obtained based on the image error between each corresponding simulated observation image and the original scan frame. For each pathological structure in the image to be reviewed and restored, positive and negative structural constraints are set. The positive structural constraints are used to constrain the corresponding pathological structure to maintain its current structural state, and the negative structural constraints are used to constrain the corresponding pathological structure to have structural modifications. Obtain the optimal positive recovery solution and the optimal negative recovery solution for each pathological structure. The optimal positive recovery solution is the candidate reconstructed image that satisfies the corresponding positive structure constraint and has the smallest robust observation distance. The optimal negative recovery solution is the candidate reconstructed image that satisfies the corresponding negative structure constraint and has the smallest robust observation distance. For any pathological structure, an evidence margin is constructed between the corresponding optimal positive recovery solution and the optimal negative recovery solution. The evidence margin represents the difference in robust observation distance between the corresponding optimal negative recovery solution and the optimal positive recovery solution. Based on the evidence margin, the review result of the corresponding pathological structure in the candidate reconstructed image is obtained.

[0017] In the current embodiment, the scan metadata is imaging data used to generate scanned raw frames. The scan metadata includes image block coordinates, objective magnification, numerical aperture, Z-axis position, focal plane prediction value, exposure start and end times, camera frame sequence, XY platform speed and acceleration, encoder trajectory during exposure, illumination power, sensor gain, sensor temperature, RAW data format, color sampling arrangement, device point spread function calibration, dark field, flat field, bad pixel and noise calibration data.

[0018] Specifically, in this scheme, the original scan frame and scan metadata are associated with a unique identifier.

[0019] In the current embodiment, the generative restoration model used in this solution can be a model for deblurring, denoising, super-resolution, digital re-aggregation, color restoration, illumination correction, or joint restoration. After obtaining the image to be restored, the model version, restoration task, restoration intensity, random seed, diffusion steps, or latent variables are recorded for subsequent calculations.

[0020] In the multiple candidate reconstruction steps of the image to be reviewed, multiple candidate reconstructed images are generated through candidate reconstruction methods such as diffusion posterior sampling, conditional normalized flow, Bayesian neural network, multiple random seed generation, latent variable optimization, or degenerate operator null space perturbation.

[0021] In the current embodiment, the calibration baseline value under each data dimension of the scan metadata is obtained, and an allowable variation range is preset for the calibration baseline value under each data dimension. Multiple sets of device degradation parameters are obtained by random sampling under each data dimension of the scan metadata based on the corresponding allowable variation range.

[0022] Specifically, the multiple sets of equipment degradation parameters are expressed as follows:

[0023] in, This represents multiple sets of device degradation parameters corresponding to the scan metadata M. This represents the total number of sets of equipment degradation parameters. This represents the degradation parameters of the Lth group of equipment.

[0024] In other words, the device degradation parameters are a set of imaging parameters that correspond one-to-one with each data dimension of the scan metadata. The data dimensions of the device degradation parameters in this scheme include optical point spread function, platform motion kernel, exposure gain, and noise parameters, etc. For example, the first... The parameters of the group of equipment are expressed as follows:

[0025] in, For the first Group equipment degradation parameters, The optical point spread function, As the core of platform movement, For exposure gain, This is a noise parameter.

[0026] Specifically, due to measurement errors in equipment calibration values, and the fact that optical state, platform movement, exposure response, and sensor noise change with scanning position, time, and operating status, the same original sensor observation may correspond to multiple sets of reasonable equipment degradation parameters, and a single set of fixed parameters cannot be used for verification.

[0027] Specifically, during the random sampling of equipment degradation parameters, priority should be given to sampling the baseline parameter group and the boundary parameter group that may lead to unfavorable audit results in order to obtain better audit results.

[0028] Specifically, parameter combinations that clearly exceed the physical performance range of the equipment or cannot explain the current scanning state are not considered equipment degradation parameters. The number L of parameter groups for equipment degradation parameters can be preset based on the equipment calibration accuracy, parameter variation range, and computing resources.

[0029] In the current embodiment, a corresponding device degradation operator is generated for each set of device degradation parameters, and a simulated observation image is obtained by simulating the original observation of the candidate reconstructed image under the corresponding device state based on the device degradation operator.

[0030] Specifically, the device degradation operator is used to simulate the process of forming the original sensor observation from a candidate sharp image under the corresponding device state, and the formula is expressed as:

[0031] in, For equipment degradation operators For candidate sharp images The simulated observation image obtained through simulation, This indicates the use of the optical point spread function for candidate sharp images. To achieve optical blurring This indicates that motion blur is applied based on the platform's movement trajectory within the exposure window. This indicates an intensity response transformation based on exposure time, illumination power, and camera gain. This indicates the sensor sampling, color response, quantization, saturation, or pixel response process.

[0032] Specifically, for spatially varying aberrations, the corresponding local point spread function is selected based on different field-of-view positions in the image, or the image is divided into multiple sub-regions and optical degradation is performed on each sub-region separately. For continuous motion scanning, the distribution of displacement in the image coordinates at each moment is calculated based on the exposure start time, exposure time, and encoder trajectory within the exposure window, and a motion kernel is formed accordingly.

[0033] Specifically, different equipment degradation parameters correspond to different equipment degradation operators. The first set of parameters can correspond to a smaller defocus amount and a shorter motion kernel, while the second set of parameters can correspond to a larger defocus amount or a longer motion kernel. The formula for obtaining multiple simulated observation images is expressed as follows:

[0034] in, For equipment degradation operators For candidate sharp images Simulated observation images obtained through simulation.

[0035] Specifically, each candidate reconstructed image is simulated using a device degradation operator, thereby simulating the multiple sensor observation results that the same clear candidate image may produce when there are reasonable fluctuations in the current device parameters.

[0036] In the current embodiment, the noise covariance corresponding to each set of device degradation parameters is obtained, and the noise whitening observation distance between each simulated observation image and the original scan frame is calculated based on the corresponding noise covariance. All noise whitening observation distances of the same simulated observation image under different device degradation parameters are integrated to obtain the robust observation distance of the corresponding candidate reconstructed image.

[0037] Specifically, in this scheme, the noise covariance is not directly incorporated into the equipment degradation operator. Instead, it is established based on the exposure, gain, predicted signal strength, and sensor calibration results corresponding to each set of equipment degradation parameters. The formula for the noise covariance is as follows:

[0038] in, The noise covariance corresponds to the equipment degradation parameters. Characterizes and predicts photon noise related to signal intensity. The sensor readout noise is characterized by the noise covariance, which describes the permissible range of random fluctuations in the corresponding simulated observation image.

[0039] Furthermore, the formula for the whitening observation distance is expressed as:

[0040] in, For noise whitening observation distance, To scan the original frame, For equipment degradation operators For candidate sharp images The simulated observation image obtained through simulation, This represents the corresponding noise covariance.

[0041] In other words, this scheme normalizes the error between the simulated observation image and the original scanned frame by using noise covariance, thereby eliminating the influence caused by different noise levels due to different degradation parameters of different devices. Finally, it integrates all noise whitening observation distances of the same simulated observation image under different device degradation parameters to obtain the robust observation distance of the corresponding candidate reconstructed image.

[0042] Specifically, a robust aggregation operator is constructed based on the maximum value, a preset high quantile, the average value of several large observation distances, or the conditional risk value. Based on the robust aggregation operator, all noise-whitened observation distances of the same simulated observation image under different device degradation parameters are integrated.

[0043] Specifically, the significance of constructing robust aggregation operators lies in the fact that if an image has a small distance only under a few specific parameter combinations, but a large distance under other equally reasonable parameter combinations, its robust distance is still relatively large, thus preventing it from passing the review based solely on a favorable set of parameters. If the restored image cannot explain the original observations under most or the most unfavorable reasonable parameters, it is deemed physically unfeasible and not allowed as the default augmentation image.

[0044] In the current embodiment, with The formula for obtaining the observationally equivalent reconstructed solution, used as the threshold for determining observation equivalence, is expressed as follows:

[0045] in, For index Robust observation distance of candidate reconstructed images.

[0046] In the current embodiment, in the step of obtaining multiple observation equivalent recovery solutions, the inter-solution feature differences are constructed as constraints, which are used to constrain the similarity of the feature vectors of any two observation equivalent recovery solutions to be less than the difference threshold.

[0047] In other words, for any two predicted observations, the equivalent reconstructed solution and Feature extraction and comparison are performed using a feature extraction function to ensure that multiple equivalent recovery solutions generated can all explain the same original sensor observations, while covering different pathological structural states as much as possible, thereby revealing structural contents that cannot be uniquely determined by the original observations.

[0048] In the current embodiment, a pathological segmentation model is used to segment each pathological structure in the image to be reviewed and restored to obtain each pathological structure in the image to be reviewed and restored.

[0049] Specifically, pathological segmentation models can be constructed using basic pathological models such as instance segmentation, gland segmentation, membrane structure detection, tissue region segmentation, and topological analysis.

[0050] In the current embodiment, a corresponding local neighborhood is preset based on the structural information of each pathological structure. The positive structural constraint is used to constrain the current structural state within the local neighborhood of the corresponding pathological structure; the negative structural constraint is used to constrain structural modifications within the local neighborhood of the corresponding pathological structure.

[0051] Specifically, based on the segmentation mask, boundaries, quantity, and connectivity of the pathological structure, corresponding local neighborhoods are preset. The constraints of the positive structure constraint include maintaining the current structural state of the pathological structure, maintaining the existence of the target structure, maintaining the structure type, maintaining the current quantity, or maintaining its boundaries and connectivity. The constraints of the negative structure constraint form at least one alternative structural state of the target structure that is different from the current state, including deleting a cell nucleus or weakening its nuclear membrane, splitting a cell nucleus into two, or fusing adjacent cell nuclei, disconnecting or connecting gland boundaries, membrane structures or blood vessel boundaries, adding or deleting local pores and lumens, reducing, adding or redistributing local staining positive areas, etc.

[0052] For example, for a single cell nucleus structure to be reviewed, a positive structural constraint requires that the nucleus continue to exist as a complete cell nucleus; a negative structural constraint may require that the nucleus be deleted or split into two nuclei. For a gland boundary to be reviewed, a positive structural constraint requires that its original connectivity remain unchanged; a negative structural constraint may require that the boundary be broken or connected to adjacent boundaries.

[0053] In other words, the biggest difference between the optimal positive recovery solution and the optimal negative recovery solution obtained by this scheme lies in the local neighborhood of the corresponding pathological structure. In other regions outside the local neighborhood, the low-frequency tissue field, illumination field and color field of the two should be consistent or the difference should not exceed the preset threshold. Furthermore, the difference between the two sensor reprojection should be within the noise allowable range. The advantage is that the margin of subsequent calculations can accurately reflect the uncertainty of the target pathological structure itself, and avoid structural changes in irrelevant regions from interfering with the final review results.

[0054] In the current embodiment, when the evidence margin is greater than the confidence threshold, the review result of the corresponding pathological structure in the candidate reconstructed image is a credible structure; when the evidence margin is not greater than the confidence threshold, the review result of the corresponding pathological structure in the candidate reconstructed image is an untrustworthy structure.

[0055] Specifically, the formula for calculating the margin of evidence is expressed as:

[0056] in, Pathological structure Evidence margin The robust observation distance for the optimal negative recovery solution. This is the robust observation distance for the optimal positive recovery solution.

[0057] Specifically, the confidence threshold in this scheme is a positive integer greater than 0. This means that since the robust observation distance characterizes the fit between the restored image and the original scan frame, a smaller value indicates a better fit. Therefore, if the robust observation distance of the optimal positive restoration solution under positive structure constraints is much smaller than the robust observation distance of the optimal negative restoration solution under negative structure constraints, a larger evidence margin can be obtained. This indicates that the fit of the current pathological structure under review to its original state is significantly better than that of the modified state, and the original structure has higher confidence. Conversely, when the evidence margin is not greater than the confidence threshold, it indicates that the original observation cannot effectively distinguish the fit between the original structure and the alternative structure, the uncertainty of the original pathological structure is high, and the review result is judged as unreliable, requiring subsequent manual verification of the structural state. Therefore, when the evidence margin is greater than the confidence threshold, it means that maintaining the current structural state can better interpret the original scan frame; when the evidence margin is not greater than the confidence threshold and is less than 0, it means that the modified pathological structure can better interpret the original scan frame, and the corresponding pathological structure may be generated by model illusion; in addition, when the evidence margin is greater than 0 and less than the confidence threshold, it means that whether the structure is modified or the current structural state is maintained, it has no impact on the interpretation of the original scan frame, so the current pathological structure is also an unreliable structure.

[0058] In addition, this scheme introduces structural topological change and model uncertainty. When the structural topological change of a pathological structure in multiple candidate reconstructed images is greater than a set threshold, or the model uncertainty is greater than a set threshold, it indicates that there is a significant discrepancy in the corresponding pathological structure, and the corresponding pathological structure is considered to be an unreliable structure.

[0059] In the current embodiment, the pathological structure with an unreliable structure as the evidence-insufficient structure is considered as the evidence-insufficient structure. Multiple supplementary sampling actions are generated based on the evidence-insufficient structure, and a corresponding physical imaging operator is generated for each supplementary sampling action. The corresponding optimal negative recovery solution is processed by each physical imaging operator to obtain multiple negative structure observation distributions, and the corresponding optimal positive recovery solution is processed by each physical imaging operator to obtain multiple positive structure observation distributions. The negative structure observation distributions and positive structure observation distributions corresponding to the same physical imaging operator are used to form observation distribution pairs. The separability between each observation distribution pair is calculated, and the supplementary sampling action corresponding to the observation distribution pair with the largest separability is taken as the supplementary sampling action to be implemented. Supplementary sampling is performed using the supplementary sampling action to be implemented. The positive structure observation distribution is the distribution of observations formed on the sensor by the optimal positive recovery solution when the corresponding supplementary sampling action is executed, and the negative structure observation distribution is the distribution of observations formed on the sensor by the optimal negative recovery solution when the corresponding supplementary sampling action is executed.

[0060] Specifically, a device action library is pre-built, and all the actions to be acquired are stored in the device action library. The re-acquisition action refers to the operation of making the scanner reacquire the original sensor observations of the area where the target structure is located. The specific re-acquisition action can be selected from the device action library according to the reason for insufficient evidence.

[0061] For example, when the lack of evidence is mainly related to defocus or missing focal layers, refocusing or adding focal layers is selected; when it is related to motion blur, reducing the platform scanning speed or acquiring a second independent frame is selected; when it is related to insufficient signal-to-noise ratio, extending the exposure time or increasing the illumination power is selected; when it is related to insufficient structural boundaries, staining response, or channel information, changing the illumination angle, wavelength, or switching the imaging channel is selected.

[0062] Specifically, the physical imaging operator is a specific form of the device degradation operator under candidate action conditions. It describes the observation that may be formed on the sensor after the candidate sharp image has undergone optical degradation, motion degradation, exposure and channel response, sensor sampling and noise superposition, following the execution of the action. For different re-acquisition actions, the corresponding physical imaging operator uses the focal layer, platform speed, exposure time, illumination or imaging channel parameters changed by the action, respectively, so as to predict "under what imaging conditions the target structure will be re-acquired if the action is executed".

[0063] Specifically, when predicting the imaging effect after performing a certain supplementary acquisition action, the optimal positive recovery solution of "preserving the target structure" and the optimal negative recovery solution of "changing the target structure" are used as inputs. After mapping by the physical imaging operator corresponding to the action, two sets of future sensor observation distributions are obtained, which are used to characterize the most likely observations to be acquired under the two structural assumptions under the action conditions.

[0064] Specifically, since both positive and negative structure observation distributions originate from the same physical imaging operator corresponding to the same re-acquisition action, they are used in pairs to examine whether sensor observations generated by the two assumptions of preserving and changing the structure can be distinguished from each other under the same re-acquisition conditions. One re-acquisition action corresponds to one pair of observation distributions, thus transforming multiple candidate actions into multiple pairs of observation distributions for comparison.

[0065] Specifically, the degree of separability measures whether two observation distributions can be distinguished. A higher degree of separability indicates that, under the supplementary acquisition action, even considering reasonable fluctuations in equipment parameters, the sensor observations generated by the two hypotheses—preserving the structure and changing the structure—can still be clearly distinguished, thus effectively eliminating structural ambiguity. The degree of separability can be calculated using worst-case Chernov information, Bach distance, or equivalent indices. Selecting the observation distribution with the highest degree of separability to perform supplementary acquisition on the corresponding action can obtain the true original observations sufficient to determine the authenticity of the structure in the most efficient way, providing a basis for subsequent verification of the structural hypothesis.

[0066] Example 2 Based on the same concept, referencing Figure 3 This application also proposes a pathological image review device based on the separability of original observations of structural counterfactuality, comprising: The acquisition module is used to acquire the original scan frame of the pathology scanner and the corresponding scan metadata, process the original scan frame using a generative recovery model to obtain the image to be recovered for review, and perform multiple candidate reconstructions on the image to be recovered to obtain multiple candidate reconstructed images. The simulation module constructs multiple sets of device degradation parameters based on scan metadata. Each candidate reconstructed image is simulated using each set of device degradation parameters to generate multiple simulated observation images. The robust observation distance of each candidate reconstructed image is calculated. Candidate reconstructed images with robust observation distances less than or equal to the observation equivalence determination threshold are taken as observation equivalence recovery solutions. For each candidate reconstructed image, the robust observation distance of the corresponding candidate reconstructed image is obtained based on the image error between each corresponding simulated observation image and the original scan frame. The constraint construction module sets positive and negative structural constraints for each pathological structure in the image to be reviewed and restored. The positive structural constraints are used to constrain the corresponding pathological structure to maintain its current structural state, and the negative structural constraints are used to constrain the corresponding pathological structure from having structural modifications. The constraint module obtains the optimal positive recovery solution and the optimal negative recovery solution for each pathological structure. The optimal positive recovery solution is the candidate reconstructed image that satisfies the corresponding positive structure constraint and has the smallest robust observation distance. The optimal negative recovery solution is the candidate reconstructed image that satisfies the corresponding negative structure constraint and has the smallest robust observation distance. The review module constructs an evidence margin between the corresponding optimal positive recovery solution and the optimal negative recovery solution for any pathological structure. The evidence margin represents the difference in robust observation distance between the corresponding optimal negative recovery solution and the optimal positive recovery solution. Based on the evidence margin, the review result of the corresponding pathological structure in the candidate reconstructed image is obtained.

[0067] Example 3 This embodiment also provides an electronic device, see reference. Figure 3 It includes a memory 402 and a processor 401, wherein the memory 402 stores a computer program and the processor 401 is configured to run the computer program to perform the steps in any of the above method embodiments.

[0068] Specifically, the processor 401 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0069] The memory 402 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 402 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 402 may include removable or non-removable (or fixed) media. Where appropriate, the memory 402 may be internal or external to a data processing device. In a particular embodiment, the memory 402 is non-volatile memory. In a particular embodiment, the memory 402 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0070] The memory 402 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 401.

[0071] The processor 401 reads and executes computer program instructions stored in the memory 402 to implement any of the pathological image review methods based on the separability of original observations of structural counterfactuals in the above embodiments.

[0072] Optionally, the electronic device may further include a transmission device 403 and an input / output device 404, wherein the transmission device 403 is connected to the processor 401 and the input / output device 404 is connected to the processor 401.

[0073] The transmission device 403 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 403 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0074] Input / output device 404 is used to input or output information. In this embodiment, the input information may be the original scan frame and corresponding scan metadata, etc., and the output information may be the review results of each pathological structure, etc.

[0075] Optionally, in this embodiment, the processor 401 can be configured to perform the following steps via a computer program: The original scan frame and corresponding scan metadata of the pathology scanner are obtained. The original scan frame is processed using a generative recovery model to obtain the image to be recovered for review. The image to be recovered for review is then reconstructed multiple times to obtain multiple candidate reconstructed images. Multiple sets of device degradation parameters are constructed based on scan metadata. Each candidate reconstructed image is simulated using each set of device degradation parameters to generate multiple simulated observation images. The robust observation distance of each candidate reconstructed image is calculated. Candidate reconstructed images with robust observation distance less than or equal to the observation equivalence determination threshold are taken as observation equivalence recovery solutions. For each candidate reconstructed image, the robust observation distance of the corresponding candidate reconstructed image is obtained based on the image error between each corresponding simulated observation image and the original scan frame. For each pathological structure in the image to be reviewed and restored, positive and negative structural constraints are set. The positive structural constraints are used to constrain the corresponding pathological structure to maintain its current structural state, and the negative structural constraints are used to constrain the corresponding pathological structure to have structural modifications. Obtain the optimal positive recovery solution and the optimal negative recovery solution for each pathological structure. The optimal positive recovery solution is the candidate reconstructed image that satisfies the corresponding positive structure constraint and has the smallest robust observation distance. The optimal negative recovery solution is the candidate reconstructed image that satisfies the corresponding negative structure constraint and has the smallest robust observation distance. For any pathological structure, an evidence margin is constructed between the corresponding optimal positive recovery solution and the optimal negative recovery solution. The evidence margin represents the difference in robust observation distance between the corresponding optimal negative recovery solution and the optimal positive recovery solution. Based on the evidence margin, the review result of the corresponding pathological structure in the candidate reconstructed image is obtained.

[0076] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0077] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0078] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets, and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product may include one or more computer-executable components configured to perform the embodiments when the program is run. The one or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted in this respect that, as Figure 3Any box in the logical flow can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0079] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for reviewing pathological images based on the separability of original observations in a structural counterfactual manner, characterized in that, Includes the following steps: The original scan frame and corresponding scan metadata of the pathology scanner are obtained. The original scan frame is processed using a generative recovery model to obtain the image to be recovered for review. The image to be recovered for review is then reconstructed multiple times to obtain multiple candidate reconstructed images. Multiple sets of device degradation parameters are constructed based on scan metadata. Each candidate reconstructed image is simulated using each set of device degradation parameters to generate multiple simulated observation images. The robust observation distance of each candidate reconstructed image is calculated. Candidate reconstructed images with robust observation distance less than or equal to the observation equivalence determination threshold are taken as observation equivalence recovery solutions. For each candidate reconstructed image, the robust observation distance of the corresponding candidate reconstructed image is obtained based on the image error between each corresponding simulated observation image and the original scan frame. For each pathological structure in the image to be reviewed and restored, positive and negative structural constraints are set. The positive structural constraints are used to constrain the corresponding pathological structure to maintain its current structural state, and the negative structural constraints are used to constrain the corresponding pathological structure to have structural modifications. Obtain the optimal positive recovery solution and the optimal negative recovery solution for each pathological structure. The optimal positive recovery solution is the candidate reconstructed image that satisfies the corresponding positive structure constraint and has the smallest robust observation distance. The optimal negative recovery solution is the candidate reconstructed image that satisfies the corresponding negative structure constraint and has the smallest robust observation distance. For any pathological structure, an evidence margin is constructed between the corresponding optimal positive recovery solution and the optimal negative recovery solution. The evidence margin represents the difference in robust observation distance between the corresponding optimal negative recovery solution and the optimal positive recovery solution. Based on the evidence margin, the review result of the corresponding pathological structure in the candidate reconstructed image is obtained.

2. The method for reviewing pathological images based on the separability of original observations according to claim 1, characterized in that, Obtain the calibration baseline value for each data dimension of the scan metadata, preset the allowable variation range for the calibration baseline value for each data dimension, and randomly sample multiple sets of device degradation parameters based on the corresponding allowable variation range for each data dimension of the scan metadata.

3. The method for reviewing pathological images based on the separability of original observations according to claim 1, characterized in that, Obtain the noise covariance corresponding to each set of device degradation parameters, calculate the noise whitening observation distance between each simulated observation image and the original scan frame based on the corresponding noise covariance, and integrate all the noise whitening observation distances of the same simulated observation image under different device degradation parameters to obtain the robust observation distance of the corresponding candidate reconstructed image.

4. The method for reviewing pathological images based on the separability of original observations according to claim 3, characterized in that, The formula for the whitening observation distance is expressed as: in, For noise whitening observation distance, To scan the original frame, For equipment degradation operators For candidate sharp images The simulated observation image obtained through simulation, This represents the corresponding noise covariance.

5. A method for reviewing pathological images based on the separability of original observations according to claim 1, characterized in that, In the step of obtaining the observation equivalent recovery solution, the feature difference between solutions is constructed as a constraint condition. The feature difference between solutions is used to constrain the similarity of the feature vectors of any two observation equivalent recovery solutions to be less than the difference threshold.

6. The method for reviewing pathological images based on the separability of original observations according to claim 1, characterized in that, Based on the structural information of each pathological structure, a corresponding local neighborhood is preset. The positive structural constraint is used to constrain the current structural state within the local neighborhood of the corresponding pathological structure; the negative structural constraint is used to constrain structural modifications within the local neighborhood of the corresponding pathological structure.

7. A method for reviewing pathological images based on the separability of original observations according to claim 1, characterized in that, When the evidence margin is greater than the confidence threshold, the review result of the corresponding pathological structure in the candidate reconstructed image is a credible structure; when the evidence margin is not greater than the confidence threshold, the review result of the corresponding pathological structure in the candidate reconstructed image is an untrustworthy structure.

8. A method for reviewing pathological images based on the separability of original observations according to claim 1, characterized in that, Pathological structures deemed unreliable by the review results are considered as insufficient evidence structures. Multiple supplementary sampling actions are generated based on these insufficient evidence structures, and a corresponding physical imaging operator is generated for each supplementary sampling action. The corresponding optimal negative recovery solution is processed by each physical imaging operator to obtain multiple negative structure observation distributions, and the corresponding optimal positive recovery solution is processed by each physical imaging operator to obtain multiple positive structure observation distributions. Observation distribution pairs are formed by negative and positive structure observation distributions corresponding to the same physical imaging operator. The separability between each observation distribution pair is calculated, and the supplementary sampling action corresponding to the observation distribution pair with the highest separability is selected as the supplementary sampling action to be implemented. Supplementary sampling is performed using the supplementary sampling action to be implemented. The positive structure observation distribution is the distribution of observations formed on the sensor by the optimal positive recovery solution when the corresponding supplementary sampling action is executed, and the negative structure observation distribution is the distribution of observations formed on the sensor by the optimal negative recovery solution when the corresponding supplementary sampling action is executed.

9. A pathological image review device based on the separability of original observations of structural counterfactuality, characterized in that, include: The acquisition module is used to acquire the original scan frame of the pathology scanner and the corresponding scan metadata, process the original scan frame using a generative recovery model to obtain the image to be recovered for review, and perform multiple candidate reconstructions on the image to be recovered to obtain multiple candidate reconstructed images. The simulation module constructs multiple sets of device degradation parameters based on scan metadata. Each candidate reconstructed image is simulated using each set of device degradation parameters to generate multiple simulated observation images. The robust observation distance of each candidate reconstructed image is calculated. Candidate reconstructed images with robust observation distances less than or equal to the observation equivalence determination threshold are taken as observation equivalence recovery solutions. For each candidate reconstructed image, the robust observation distance of the corresponding candidate reconstructed image is obtained based on the image error between each corresponding simulated observation image and the original scan frame. The constraint construction module sets positive and negative structural constraints for each pathological structure in the image to be reviewed and restored. The positive structural constraints are used to constrain the corresponding pathological structure to maintain its current structural state, and the negative structural constraints are used to constrain the corresponding pathological structure from having structural modifications. The constraint module obtains the optimal positive recovery solution and the optimal negative recovery solution for each pathological structure. The optimal positive recovery solution is the candidate reconstructed image that satisfies the corresponding positive structure constraint and has the smallest robust observation distance. The optimal negative recovery solution is the candidate reconstructed image that satisfies the corresponding negative structure constraint and has the smallest robust observation distance. The review module constructs an evidence margin between the corresponding optimal positive recovery solution and the optimal negative recovery solution for any pathological structure. The evidence margin represents the difference in robust observation distance between the corresponding optimal negative recovery solution and the optimal positive recovery solution. Based on the evidence margin, the review result of the corresponding pathological structure in the candidate reconstructed image is obtained.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform a pathological image review method based on the separability of original observations of structural counterfactuals as described in any one of claims 1-8.

Images