Digital pathology scanner color variation automated detection and correction method, apparatus and media

Abstract

The application provides a digital pathology scanner color change automatic detection and correction method, device and medium, aiming at the problem of periodic dependence on physical color card for color drift detection, a one-time reference calibration is performed on a standard color card after installation or overhaul; a self-reference detection system including dark field frame zero reference, background colorimetric reference and tile overlap area verification is constructed in daily operation, the detection of color transfer function component drift is decoupled and three-level abnormal grading is performed; for normal or interpretable abnormality, correction parameters are dynamically generated based on the degradation prediction model and the self-reference measurement value, and color compensation is applied; only when the abnormality is not interpretable, the color card bottom verification is triggered. It is mainly used for color consistency guarantee in digital pathology section scanning, remote consultation and image quantitative analysis.

Description

Technical Field

[0001] This invention relates to the field of digital pathological imaging and color management technology, and in particular to an automated detection and correction method, device and medium for color changes in digital pathological scanners, which is especially suitable for automatic monitoring and color deviation compensation of the scanner's color transfer function in scenarios such as digital scanning of pathological slides, remote pathological consultation and artificial intelligence-assisted pathological diagnosis. Background Technology

[0002] Digital pathology scanners, through sophisticated optical systems and high-resolution linear or area array sensors, transform tissue specimens on glass slides into whole-slide digital images that pathologists can view on a monitor. In routine pathological procedures such as hematoxylin-eosin (H&E) staining and immunohistochemical (IHC) staining, accurate color reproduction directly impacts the reliability of diagnostic conclusions. Taking H&E staining as an example, the difference in hue and saturation between the blue of hematoxylin (cell nucleus) and the pink of eosin (cytoplasm) is a crucial basis for pathologists' judgment. When the scanner's light source color temperature shifts from 5500K to 5200K, the blue channel response decreases by approximately 5%, causing eosin pink to lean towards dark red and hematoxylin blue towards bluish-gray. While this change is visually perceptible, it is easily overlooked in the absence of objective quantitative indicators.

[0003] Currently, the industry generally uses standard color charts (such as the IT8.7 / 1 color chart) as the benchmark reference for color management and calibration. This involves inserting the color chart into the scanner during installation, debugging, and periodic maintenance. By scanning color patches with known spectral reflectance characteristics, the scanner's color transfer function is measured, and calibration parameters are established. Other solutions incorporate a white reference plate or blank hue reference within the scanner. Before each scan, a reference signal is acquired for white balance or flat field correction to suppress the effects of lighting variations.

[0004] However, the aforementioned periodic color chart calibration method has inherent limitations: First, color chart calibration is a discrete, offline process, and gradual color shifts occurring during the operational period between calibrations cannot be detected in a timely manner, resulting in monitoring blind spots lasting for weeks or even months. Second, the dye layer of the color chart will slowly fade under the influence of light, temperature, and humidity, and the drift of the color chart's own reference value introduces secondary errors, with the decay rate accelerating with increased usage frequency. Third, each color chart calibration requires manual insertion of the color chart and a 15-30 minute shutdown, affecting clinical testing throughput, and the cost of color chart consumables is high. Fourth, the equipment status at the time of calibration may not represent the typical status during the operational period; the light source and sensor status are affected by dynamic factors such as temperature and continuous operating time, and the calibration results at a single point in time cannot reflect the dynamic evolution during the operational period. Furthermore, the solution with a built-in white reference plate can only correct for changes in illumination brightness and cannot distinguish color changes caused by different physical mechanisms such as sensor dark current drift, optical coating oxidation, and electronic gain drift. It also lacks the ability to locate and classify the root causes of color shifts, making it difficult to meet the stringent requirements of long-term color consistency in pathological diagnosis.

[0005] Therefore, there is an urgent need for an automated method, device, and medium for detecting and correcting color changes in digital pathology scanners to solve the problems existing in the current technology. Summary of the Invention

[0006] This invention provides an automated detection and correction method, device, and medium for color changes in digital pathology scanners. It addresses the problems of existing periodic discrete calibration methods based on standard color cards, which cannot capture the continuous dynamic drift of the scanner's color transfer function between two calibrations in real time, resulting in a monitoring blind spot lasting for weeks or even months. These methods cannot achieve early warning and proactive intervention for color deviation, and they also have a continuous dependence on manual operation and color card consumables.

[0007] The core technology of this invention is to construct a three-dimensional orthogonal self-reference color change detection system consisting of a dark field frame zero reference, a background chromaticity ensemble reference, and a consistency check of the overlapping area of ​​scanned tiles. By decomposing the color transfer function into independently monitorable physical components, the root cause of drift is accurately located. A predictive correction mechanism based on the fusion of an exponential decay model and Kalman filtering is used to achieve feedforward compensation and active correction of color cast. The standard color card is only triggered for fallback verification when an inexplicable anomaly occurs.

[0008] In a first aspect, the present invention provides an automated detection and correction method for color changes in a digital pathology scanner, the method comprising the following steps: During scanner installation or after major overhaul, use a standard color chart to perform baseline color calibration and obtain initial color characterization parameters and baseline reference data. During the daily operation of the scanner, a self-reference detection system is constructed, which includes a dark field frame zero reference dimension, a background chromaticity ensemble reference dimension, and a spatial redundancy check dimension. Each dimension is independent of the others and is redundant to each other. Based on the self-reference signal acquired in real time by the self-reference detection system, the drift state of different physical components in the color transfer function of the detection scanner is decoupled and the detected drift state is divided into normal state, explainable abnormal state and unexplainable abnormal state according to the preset rules. In response to whether the detected drift state is normal or an interpretable abnormal state, correction parameters are dynamically generated based on the degradation prediction model and the measured values ​​of the self-reference signal, and color compensation is applied to the scanner. In response to the detected drift state being an unexplained anomaly, automatic correction is paused and full color gamut verification of the standard color chart is triggered.

[0009] Furthermore, the zero-reference dimension of the dark frame includes: Before scanning each batch of slices, the light path is blocked by at least one of the following methods: turning off the light source, closing the shutter, or blocking the light path. Multiple dark field images are acquired and the average value is calculated. The average pixel value of the dark field image in each color channel is calculated to obtain the current dark current. Based on the real-time operating temperature of the sensor, a preset dark current temperature index model is input to verify the conformity of the current dark current with temperature drift, so as to distinguish between normal temperature drift that conforms to the temperature model and dark current anomalies that exceed the model prediction. During image generation, baseline correction is performed by subtracting the current dark current of the corresponding channel from the original count value of each pixel.

[0010] Furthermore, the background chromaticity ensemble reference dimensions include: Foreground and background segmentation is performed in the low-resolution pyramid layer of the sliced ​​panoramic image, and a tissue binary mask is generated and upsampled to the analysis resolution layer. Based on the area threshold, distance constraints from the tissue boundary, and color distribution variance within the region, valid background candidate regions are selected. Pixel sampling is performed within the valid background candidate region, and the truncated mean is calculated to obtain the background representative color of the slice. Ensemble statistical modeling is performed on the background representative colors of multiple slices within a single batch. Normalized chromaticity coordinates and channel mean values ​​are calculated. The fluctuation of the batch background mean value converges as the sample size increases, serving as a stable virtual reference. The type of spectral drift of the light source can be identified by the direction and magnitude of the offset of the normalized chromaticity coordinates relative to the reference state.

[0011] Furthermore, the drift states of different physical components in the color transfer function of the decoupled detection scanner include: The color transfer function is decomposed into dark current component, light source and optical transmittance combined spectral component, electronic gain component, and spatial uniformity component. Among them, the dark current component is monitored by the zero reference dimension of the dark field frame, the joint spectral component is monitored by the background chromaticity ensemble reference dimension, the electronic gain component is monitored by the joint analysis of the dark field frame and the background chromaticity, and the spatial uniformity component is monitored by the spatial redundancy check dimension. In the spatial redundancy check dimension, the pixel pairs of adjacent tiles at the same physical location in the overlapping area are extracted according to the scanning tile layout parameters, the pixel difference statistics of each color channel are calculated, and orthogonal cross-validation is performed with the detection results of the dark field frame zero reference dimension and the background chromaticity ensemble reference dimension. When an anomaly is detected in only a single dimension, the system issues an alert and continues monitoring; when anomalies occur in at least two dimensions simultaneously, it determines that the color transfer function has undergone systematic degradation and initiates a correction decision.

[0012] Furthermore, correction parameters are dynamically generated based on the degradation prediction model and the measured values ​​of the self-reference signal, and color compensation is applied to the scanner, including: After deducting the effects of the current dark current and the initial reference dark current, calculate the measured values ​​of the channel gain change for each color channel; The gain prediction value is obtained based on the exponential attenuation prediction model of light source attenuation. The gain prediction value and the channel gain change measurement value are dynamically fused and estimated by Kalman filtering algorithm to obtain the fused gain coefficient of each channel. A color correction matrix can be constructed based on the fusion gain coefficient to perform digital correction on the scanned image data, or the gain adjustment parameters obtained based on the fusion gain coefficient can be directly written into the scanner's hardware register.

[0013] Furthermore, based on preset rules, the detected drift states are divided into normal states, explainable anomalous states, and unexplainable anomalous states, including: The batch self-reference statistical sequence is monitored using statistical process control methods, including Shewhart control charts for monitoring sudden large shifts, cumulative sum control charts for monitoring small and persistent drifts, and exponentially weighted moving average control charts for monitoring mean smoothing trends. When the drift is within the confidence interval of the degradation prediction model and all control charts are in a controlled state, it is judged as a normal state; When the drift exceeds the controlled threshold, but the drift direction and evolution rate are consistent with the known physical degradation model, it is determined to be an interpretable anomalous state. When the drift amount exhibits a non-temperature-related step change, or when the drift direction is orthogonal to the historical evolution trajectory, it is determined to be an unexplainable abnormal state.

[0014] Furthermore, after applying color compensation to the scanner, the process also includes calibration parameter verification and rollback steps: Obtain the color data of the reference slice in the compensated image, and monitor the background chromaticity convergence status of subsequent scan batches after applying compensation; The calibration verification is deemed to have failed when the color deviation of the reference slice exceeds the preset tolerance range or the background chromaticity of subsequent scan batches does not return to the reference range. In response to a calibration verification failure, the system automatically rolls back to the previous set of valid calibration parameters, records the calibration log, and triggers full color gamut verification of the standard color chart.

[0015] Secondly, the present invention provides an automated detection and correction device for color changes in a digital pathology scanner, comprising: The initial calibration module is used to perform reference color calibration using a standard color chart during scanner installation or after major repair, and to obtain initial color characteristic parameters and reference data. The self-reference detection module is used to build a self-reference detection system during the daily operation of the scanner, which includes a dark field frame zero reference unit, a background chromaticity ensemble reference unit, and a spatial redundancy check unit. Each unit is independent of the others and is redundant to each other. The anomaly classification module is used to decouple the drift state of different physical components in the color transfer function of the scanner based on the self-reference signal acquired in real time by the self-reference detection system, and classify the detected drift state into normal state, explainable anomaly state and unexplainable anomaly state according to preset rules. The automatic correction module is used to dynamically generate correction parameters based on the degradation prediction model and the measured values ​​of the self-reference signal in response to whether the detected drift state is normal or an interpretable abnormal state, and to apply color compensation to the scanner. The fallback trigger module is used to pause automatic correction and trigger full color gamut verification of the standard color chart in response to the detected drift state being an unexplainable abnormal state.

[0016] Thirdly, the present invention provides 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 execute the above-described automated detection and correction method for color changes in a digital pathology scanner.

[0017] Fourthly, the present invention provides a readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for automated detection and correction of color changes in a digital pathology scanner.

[0018] The main contributions and innovations of this invention are as follows: 1. Dark field frame zero reference dimension: Before scanning each batch, the light path is blocked to acquire dark field images. Background chromaticity ensemble reference dimension: After scanning each slice, the blank background area is automatically extracted for batch statistics. Spatial redundancy check dimension: During the scanning process, the overlapping area of ​​adjacent tiles is used for consistency check. All three dimensions rely on the existing data in the conventional scanning process and the existing mechanism of the scanner to run in parallel without changing the existing scanning process or adding extra scanning steps. The time resolution of color change detection is improved from the weekly or monthly level of traditional periodic color card calibration to the batch level and slice level. Gradual color shift that occurs during the operation period between two maintenance can be detected in a timely manner.

[0019] 2. Traditional methods require color card calibration every 2 to 4 weeks. This invention reduces the usage frequency of standard color cards to only once after installation or major overhaul. Subsequent verification is only triggered in the event of unexplained anomalies. The actual frequency can reach monthly or longer, directly saving the cost of color card consumables and the downtime of 15 to 30 minutes each time. At the same time, the reduced usage frequency of color cards slows down the secondary error introduced by the fading of the dye layer of the color card itself.

[0020] 3. The three dimensions of dark field frame, background chromaticity statistics, and tile overlap area monitor different physical components of the color transfer function, which are independent and redundant to each other. Anomalies in any dimension can be cross-verified by the other two dimensions: when only one dimension detects an anomaly, the system issues an early warning and continues monitoring, waiting for confirmation from subsequent batch data; when at least two dimensions show anomalies simultaneously, a systemic degradation is determined and a correction decision is initiated, avoiding misjudgments and miscorrections caused by specific interference factors such as dust or changes in slide batches affecting a single signal.

[0021] 4. The color transfer function is decomposed into dark current component, light source and optical transmittance joint spectral component, electronic gain component and spatial uniformity component. An independent detection path is assigned to each type of component, which enables the system to automatically distinguish different degradation mechanisms such as sensor aging, light source spectral drift, optical coating oxidation, and focus micro-drift. Based on this, explainable and unexplainable anomalies can be distinguished, providing a diagnostic basis for graded handling and precise equipment operation and maintenance.

[0022] 5. Based on the exponential decay prediction model of light source attenuation, the predicted value of gain is obtained. The predicted value and the measured value are dynamically fused by Kalman filtering. This can predict the gain change and apply compensation before the color cast accumulates to a perceptible level. The correction mode is upgraded from passive feedback to active feedforward, which further reduces the color fluctuation between batches. At the same time, the channel gain change measurement value is calculated by subtracting the influence of the current dark current and the initial reference dark current, so as to avoid the confusion between the increase of dark current and gain drift.

[0023] 6. Only abnormal triggers can be explained by rapid correction. Normal drift is automatically absorbed by the prediction model without generating additional correction actions, thus avoiding oscillations introduced by frequent corrections. The correction parameters can only remain effective after being verified by comparison with the reference slice and historical benchmark, and after the background color convergence monitoring of subsequent batches passes. If the verification fails, it will automatically roll back to the previous set of valid correction parameters and record the correction log to ensure that the correction action will not introduce new color deviations.

[0024] 4. The predictive model and calibration parameters can be extended to collaborative management scenarios of multiple scanners, providing a unified color benchmark for multi-center pathological research and remote consultation; the output signal of the three-dimensional self-reference detection can also be used for equipment health diagnosis, and the continuously recorded time series data of each dimension forms a complete equipment color health profile, providing a data foundation for long-term operation and maintenance and continuous optimization of degradation models.

[0025] Details of one or more embodiments of the present invention are set forth in the following drawings and description, so that other features, objects and advantages of the invention will be more readily understood. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall architecture and data flow of an automated detection and correction system for color changes in a digital pathology scanner, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the correspondence between the root cause of color change and the detection path, provided in an embodiment of the present invention. Figure 3 This is a schematic flowchart of a dark field frame zero-reference detection method provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the workflow of a three-dimensional self-reference detection system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an anomaly classification and correction decision-making process provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of a prediction model and Kalman filter compensation data flow provided in an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the evolution of a full lifecycle workflow stage as provided in an embodiment of the present invention. Detailed Implementation

[0027] 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.

[0028] 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.

[0029] Example 1: System Overall Architecture and Functional Module Division like Figure 1 As shown, this embodiment provides an automated color change detection and correction system for a digital pathology scanner, comprising an initial calibration module, a self-reference detection module, an anomaly grading module, an automatic correction module, and a fallback trigger module. Each module can be implemented in software on the scanner's control computer, or it can be partially integrated into the scanner's embedded controller.

[0030] Initial calibration module: Used after the digital pathology scanner has been installed and debugged, or after undergoing a major optical and hardware overhaul, to perform a one-time full-gamut absolute calibration using a standard color chart (such as the IT8.7 / 1 standard color chart, or other metrologically traceable standard color charts). This calibration process is executed after the system has warmed up and reached thermal equilibrium, establishing an initial color characterization matrix. The reference color values ​​of 288 standard color patches were recorded, and background area color statistics and dark field data were collected from 50 to 100 representative slides (covering conventional staining types such as hematoxylin-eosin H&E and immunohistochemical IHC) to construct the initial parameters for the background baseline vector, dark current baseline, and statistical process control chart. After calibration, the standard color cards were retrieved and stored in the verification library, and were no longer periodically inserted for use during daily operation.

[0031] Self-reference detection module: This module constructs a self-reference detection system during the scanner's daily operation, comprising a dark-field frame zero-reference unit, a background chromaticity ensemble reference unit, and a spatial redundancy check unit. Each unit is independent and redundant, and their data is independent of each other. The three units monitor changes in the scanner's color transfer function from three independent physical dimensions. Dark field frame zero reference unit: By using the scanner's built-in shutter mechanism, light shield, or light source control circuit to block the light path, dark field images of the sensor are acquired before each batch (e.g., 500 slides) is scanned, dark current is directly extracted and temperature drift is checked, providing a physical absolute zero reference.

[0032] Background chromaticity ensemble reference unit: Automatically segments and filters the background area of ​​blank slides from the pyramid structure of the whole slide image (WSI) generated by conventional scanning, calculates the mean value and normalized chromaticity coordinates of each color channel through large-sample sampling, and provides ensemble statistical reference.

[0033] Spatial redundancy check unit: Utilizing the pixel-level overlap area between adjacent scanning tiles during linear or area array scanning imaging, it calculates the pixel difference of the same physical location in two independent acquisitions, providing a reference for spatial imaging uniformity and focus consistency.

[0034] Anomaly Classification Module: Receives the multidimensional detection signal output from the self-reference detection module, decouples and analyzes the changes in each physical component of the color transfer function, and combines the statistical process control chart and the confidence interval of the degradation prediction model to classify the color drift state into three levels: normal state, explainable abnormal state, and unexplainable abnormal state.

[0035] Automatic correction module: For normal and explainable abnormal states, based on the exponential decay prediction model fitted to historical degradation data and the Kalman filter algorithm, it dynamically fuses and estimates the current channel gain coefficient, constructs a correction matrix, applies it to image data or directly writes it to the scanner hardware gain register, and realizes active color feedforward compensation without the participation of physical color cards.

[0036] The fallback trigger module is activated only when an unexplainable abnormal state is detected. It suspends the system's automated calibration process, issues an alarm, and guides the operator to insert a standard color chart to perform full color gamut verification and recalibration.

[0037] like Figure 1 As shown, the five modules form a closed-loop data flow: the reference parameters established by the initial calibration module are input to the self-reference detection module and the automatic correction module; the self-reference detection module outputs a signal to the anomaly classification module; the anomaly classification module outputs the judgment results to the automatic correction module (normal state or interpretable anomaly state) and the fallback trigger module (uninterpretable anomaly state); the color card verification results of the fallback trigger module are used to update the reference parameters of the initial calibration module.

[0038] Example 2: Color Transfer Function Decomposition and Physical Degradation Mechanism like Figure 2 As shown, the digital pathology scanner is in time For wavelength The color transfer function of an imaging system can be modeled as:

[0039] in, For color channel indexes; For channel At any moment For wavelength The overall imaging response; For the spectral components of the light source; For the transmittance component of the optical system; Quantum efficiency for image sensors; This is the electronic gain component; For channel Dark current output.

[0040] The essence of systematic color shift is that the above components change over time. Asymmetric changes occur, causing the relative response ratios of the RGB three channels to deviate from the initial calibration state. This invention maps each physical degradation root cause to an independent colorless card detection path: LED phosphor thermal degradation: Under long-term high-power excitation, LED phosphors undergo photothermal aging, altering the blue-yellow power ratio of the emission spectrum and causing a slow increase in color temperature (typically a drift of 50 to 200 K per year). The red channel gain relatively decreases, while the blue channel gain relatively increases. This physical degradation is reflected in the spectral components of the light source. Monitoring is performed using the background chromaticity ensemble reference dimension.

[0041] Sensor dark current temperature drift: The dark current generated by thermal excitation inside the sensor chip increases exponentially with increasing temperature (the dark current doubles for every 6 to 8 degrees Celsius increase in temperature), causing a synchronous rise in the baseline of the three image channels. This degradation is mapped to the dark current component. It is extracted and verified directly through the zero-reference dimension of the dark field frame.

[0042] Oxidation of optical component coatings: Antireflective coatings on filters and lenses slowly oxidize under prolonged exposure to ultraviolet and visible light, resulting in non-uniform changes in the transmittance spectrum (with more significant attenuation in the blue light band). This degradation is reflected in the optical transmittance components. Monitoring is conducted through the background chromaticity ensemble reference dimension.

[0043] Electronic gain parameter drift: Aging of analog amplifier circuit components causes proportional misalignment in the three-channel relative response. This degradation is mapped to the electronic gain component. Joint monitoring is performed by measuring the slope changes of the dark field baseline and the background mean.

[0044] Mechanical focusing micro-drift and spatial non-uniformity: Wear of the mechanical guide rails, thermal expansion and contraction, or dust contamination lead to a decrease in image sharpness and edge illumination uniformity. This degradation is mapped to the spatial uniformity component. Monitoring is performed through the consistency verification dimension of tile overlap area.

[0045] like Figure 4 As shown in Table 1, the correspondence between each component and the monitoring path is as follows: Table 1 Mapping Table of Transfer Function Components and Monitoring Paths

[0046] Example 3: Zero-Reference Detection Method for Dark Frames like Figure 3 As shown, the specific implementation process of zero-reference detection in dark frames includes: Step 1: Dark-field frame acquisition. Before scanning each batch of slices, the light path is blocked by at least one of the following methods: turning off the light source, closing the shutter, or blocking the light path. For scanners equipped with an electric shutter mechanism, closing the shutter is preferred; for scanners without a shutter but supporting software control of the light source, turning off the light source is used; for scanners without either of these, a light-blocking plate can be inserted into the light path. With no external light incident on the sensor, 10 consecutive dark-field images are acquired and the average value is calculated to suppress random noise. Simultaneously, the real-time operating temperature of the sensor during acquisition is read and recorded using a temperature sensor. .

[0047] Step 2: Dark Current Extraction and Spatial Variance Evaluation. Calculate the mean value of all pixels in each channel for the averaged dark-field frame to obtain the current measured dark current value for each channel. ,in Simultaneously, the spatial distribution variance of dark current pixels across the entire sensor is calculated to assess the variation in sensor pixel response non-uniformity.

[0048] Step 3: Temperature Physical Model Tracking and Verification. The sensor's dark current changes with temperature following an exponential relationship model:

[0049] in, Reference temperature for initial absolute calibration The initial dark current measured below; This is the current measured temperature; For channel The temperature doubling factor can be read from the image sensor chip datasheet, with typical values ​​(usually 6 to 8 degrees Celsius) or calibrated during installation through a temperature step test—that is, controlling the sensor to acquire dark field frames at different temperatures and fitting an exponential relationship.

[0050] Step 4: Baseline subtraction correction. Obtain the verified current dark current. Then, during the image generation process, the current dark current of the corresponding channel is subtracted from the original count value of each pixel to complete the baseline correction and eliminate the baseline lifting effect of dark current on the imaging color.

[0051] Step 5: Anomaly Detection. The system compares the deviation between the measured dark current value and the model prediction value: if the measured value is within the range of the model prediction value... Within a certain range, the temperature drift is considered normal and conforms to the temperature model. If the dark current increases by more than 30% compared to the installation reference, or the difference in dark current increase among the three channels exceeds 15 percentage points, it is considered a sensor aging warning. If the dark current exhibits a step change and does not conform to the temperature model, it is considered an unexplained anomaly.

[0052] Example 4: Background Chromaticity Ensemble Reference Detection Method like Figure 4 As shown, the background chromaticity ensemble reference detection is based on the optical invariant properties of blank glass slides in everyday slides (the transmittance of blank glass slides is approximately flat in the visible light 400 to 700 nm band, with no selective absorption peaks). The specific implementation process is as follows: Step 201: Low-resolution pyramid segmentation. The full-slice digital image is stored in a pyramid hierarchical structure. At the lowest resolution layer (e.g., a 32x or 64x thumbnail), the image is divided into tissue regions and blank background regions using an Otsu adaptive threshold or a K-means clustering algorithm with a cluster size of 2.

[0053] Step 202: Morphological Refinement of Tissue Mask. Morphological opening operations are used to eliminate minor noise and closing operations are used to fill internal pores in the tissue to generate a binary tissue mask. The mask is then upsampled to the analysis resolution layer (e.g., 4x layer) as a spatial constraint for background extraction.

[0054] Step 203: Multi-condition filtering of valid background candidate regions. Further filtering of pixel regions marked as background by the mask is performed on the analysis layer: 203.1 Area Threshold Filtering: Removes isolated background fragments with an area less than 5000 pixels.

[0055] 203.2 Boundary distance constraint: Prioritize areas far from tissue boundaries to reduce contamination from tissue-scattered light.

[0056] 203.3 Internal uniformity test: Calculate the standard deviation of each RGB channel within each candidate region and remove non-uniform regions with excessively large standard deviations.

[0057] Step 204: Noise-resistant sampling and representative value calculation. Within each selected valid background area, 500 pixels are randomly sampled and their RGB values ​​are recorded. The 10% truncated mean is taken as the representative color of the area to suppress the influence of dust and noise.

[0058] Step 205: Large-sample ensemble statistical modeling. Using each batch (e.g., 500 slides) as a statistical period, statistical modeling is performed by accumulating the background representative colors of multiple slides within a single batch, calculating the channel mean vector, covariance matrix, and normalized chromaticity coordinates. According to the central limit theorem, when the sample size... When, the standard error of the batch background mean is calculated according to Convergence is achieved by using the mean background color of a large batch of samples as a stable virtual reference. The formula for calculating normalized chromaticity coordinates is:

[0059] in, , , These are the pixel values ​​for the red, green, and blue channels of the background area, respectively. , , These are the normalized chromaticity coordinates of the corresponding channels.

[0060] Step 206: Chromaticity Shift Pattern Analysis and Color Cast Identification. Under normal conditions, the normalized chromaticity coordinates of the background converge at the center point. (Neutral White). When the chromaticity coordinates deviate from the center point, the type of color cast is identified based on the direction of the offset: Decrease and Increased color temperature indicates LED phosphor degradation; Increase and Decrease the indicator to show a lower color temperature; A decrease in a single value indicates optical coating degradation or blue filter aging. Furthermore, a synchronous shift in channel mean suggests increased dark current; an asymmetric shift in channel mean suggests gain-type color shift; a small change in mean but a significant increase in variance suggests dust or uneven illumination; a decrease in saturation but a small change in mean suggests focus drift.

[0061] Example 5: Spatial Redundancy Check Dimension and Orthogonal Cross-Validation like Figure 4 As shown, digital pathology scanners use linear or area array sensors for scanning. There is a pixel-level overlap between adjacent scanning tiles (typically 64 to 128 pixels). The overlapping pixel values ​​from two independently acquired scans should be consistent. The specific implementation process for spatial redundancy verification is as follows: Step 301: Physical mapping of overlapping areas. Based on the scanned tile layout parameters (tile size, number of overlapping pixels, scan direction), the coordinate mapping of the overlapping areas of adjacent tiles is automatically calculated.

[0062] Step 302: Calculate the difference between overlapping pixels. Extract pixel pairs at the same physical location within the overlapping area and calculate the absolute difference across the three channels:

[0063] In the formula, , These are the pixels at the same physical location in the overlapping area of ​​two adjacent tiles. Channel value; For channel The pixel difference in the overlapping area. Statistical analysis of all overlapping areas. Mean, variance, and spatial distribution.

[0064] Step 303: Determining Spatial Uniformity and Channel Sensitivity. Under normal conditions... The mean should be less than 2DN, and the variance should be less than 1.5DN. Spatial homogeneity degradation is defined as the mean or variance exceeding a threshold. If the difference in the blue channel... The value is significantly greater than that of the red and green channels, indicating that the optical attenuation in the blue light band or the inconsistency in the response space of the sensor's blue channel is aggravated, which further corroborates the decay of the LED phosphor.

[0065] Step 304: Three-dimensional orthogonal cross-validation and staircase determination. The spatial redundancy check results are orthogonally cross-validated with the detection results of the zero-reference dimension of the dark frame and the background chromaticity ensemble reference dimension. The determination rules are shown in Table 2: Table 2. Three-dimensional orthogonal cross-validation judgment rules.

[0066] When an anomaly is detected in only a single dimension, the system issues an alert and continues monitoring, waiting for confirmation from subsequent batches of data; when anomalies occur in at least two dimensions simultaneously, it determines that the color transfer function has undergone systematic degradation and initiates a correction decision.

[0067] Example 6: Anomaly Classification Mechanism and Statistical Process Control (SPC) like Figure 5 As shown, the anomaly classification module divides the detected drift states into three levels: Normal state: The drift is within the confidence interval of the degradation prediction model and all statistical control charts are under control. Specifically, the dark current increase is within the temperature model's predicted value. Within this range, the background chromaticity coordinate offset is less than 0.003, and the tile overlap area... If the mean is less than 2DN and the trend of change conforms to the fluctuation range allowed by the prediction model, the prediction model will automatically incorporate the drift into continuous feedforward compensation without generating additional correction actions.

[0068] Explainable anomalies: The drift exceeds the controlled threshold, but the drift direction and evolution rate are consistent with known physical degradation models. Typical scenarios include: LED phosphor degradation leading to a slow increase in color temperature (blue channel gain continuously increases, red channel gain continuously decreases, and the rate conforms to an exponential decay model); dark current slowly increases with temperature (consistent with a temperature doubling model); and slow degradation of optical coatings leading to a decrease in blue channel transmittance. These anomalies are compensated for by a fast correction matrix and encrypted monitoring frequency, eliminating the need for a standard color chart.

[0069] Unexplained abnormal states: A non-temperature-related step change in drift (e.g., dark current increases by more than 50% in a single batch), simultaneous abrupt changes in all three channels that do not conform to the temperature model, or drift direction orthogonal to the historical evolution trajectory (e.g., historically a blue channel shift, suddenly a large shift in the red channel). The system will pause automatic correction, issue an alarm, and trigger full color gamut verification using the standard color chart.

[0070] Anomaly classification employs statistical process control (SPC) methods to perform multi-graph joint monitoring of batch self-reference statistical sequences: (1) Shewhart control chart: used to detect sudden large offsets, and the control center line is set with the initial batch data. Upper control limit , lower control limit ( and (The initial calibration is based on the statistical mean and standard deviation). An early warning is triggered when the batch mean exceeds the control limit.

[0071] (2) Cumulative control chart: used to detect small-amplitude persistent drift and calculate cumulative deviation:

[0072]

[0073] in, For the first Batch observations; The target mean; This is a reference value (usually half of the target offset); when the cumulative sum exceeds the decision threshold... An alert is triggered at any time.

[0074] (3) Exponentially weighted moving average control chart: used for smoothing trend monitoring and calculating smoothing statistics:

[0075] in, Smoothing factor; initial value Take the average of the initial batch statistics.

[0076] The three control charts are used in combination in a complementary manner, and the anomaly classification process is initiated when any control chart triggers an early warning.

[0077] Example 7: Gain Tracking, Degradation Prediction Model and Kalman Filter Correction Mechanism like Figure 6 As shown, the implementation process of the automatic calibration module is as follows: Step 401: Establish Installation Anchor Points. Establish a color feature matrix during installation calibration. Simultaneously, the first batch of background baseline data was collected: background channel mean. Dark current Chromaticity coordinate reference value This forms the initial anchor point for the degradation prediction model.

[0078] Step 402: Calculate the measured channel gain change after subtracting dark current. At time... Obtain the current dark current through dark field frames. Obtain the current background average value through background statistics. After deducting the effects of the current dark current and the initial reference dark current, the measured values ​​of gain change for each channel are calculated:

[0079] The dark current has been deducted from both the numerator and denominator, reflecting only the change in spectral response, ensuring that the increase in dark current will not be confused with the change in gain.

[0080] Step 403: Exponential decay trend prediction. The exponential decay prediction model based on light source attenuation performs least-squares fitting of historical data to the changes in the gain coefficients of each channel over time:

[0081] in, For channel The total attenuation; The decay time constant; This represents the runtime since installation. After fitting the parameters, the gain coefficient for the next batch of scans is predicted. ( (For batch intervals), feedforward compensation is applied in advance.

[0082] Step 404: Dynamic Fusion Estimation using Kalman Filtering. The discrete Kalman filter algorithm is used to dynamically fuse and estimate the predicted gain value with the measured channel gain change value.

[0083] in, These are the estimated values ​​of the fused gain coefficients; This is a predicted value; The measured value calculated in step 402 ; This is the Kalman gain. When the trend is stable... Smaller, relying more on forecasts to reduce the impact of noise; when trends change abruptly The response time is relatively large, relying more on actual testing for a rapid response.

[0084] Step 405: Correction parameter construction and dual-track application. Based on fusion gain coefficient. , , Construct a color correction matrix:

[0085] in, Represents a diagonal matrix. Color compensation supports dual-track execution mode: (1) Digital matrix correction: The color correction matrix is ​​applied to the scanned image data, and the corrected pixels are... ( (for the pixel vector before correction).

[0086] (2) Direct writing to hardware registers: When the scanner supports digitally controllable exposure time and channel gain registers, the gain adjustment parameters obtained based on the fusion gain coefficient conversion are directly written into the scanner's hardware registers to correct color cast at the source and avoid quantization errors introduced by post-processing. The two methods can be used alone or in combination, and no physical color card is used in the entire compensation process.

[0087] Example 8: Calibration Parameter Verification, Security Rollback, and Audit Logs After applying color compensation, the system performs closed-loop verification and a safety rollback: Step 501: Reference Slide Verification. Immediately after calibration, scan the archived standard H&E slide pre-placed in the quality control compartment—if the system is equipped with an autosampler, the autosampler will move it from the quality control clamp; if the system is not equipped with an autosampler, the operator will place it manually—extract the color values ​​of the characteristic regions and compare them with historical benchmarks to calculate the CIEDE2000 color difference. .

[0088] Step 502: Subsequent Batch Background Convergence Monitoring. Continuously monitor whether the background normalized chromaticity coordinates of subsequent scan batches return to the reference range after the correction is applied.

[0089] Step 503: Verification and Automatic Rollback. If the color deviation of the reference slice exceeds the preset tolerance range, or if the background chromaticity of subsequent scan batches fails to return to the baseline range, the calibration verification is deemed a failure. The system automatically cancels the current parameters, seamlessly rolls back to the previous set of valid calibration parameters, records the calibration log, and triggers full-gamut verification of the standard color chart. All calibration parameters, along with timestamps, trigger reasons, and monitoring data snapshots, are stored in the calibration log archive.

[0090] Example 9: Evolution of the Full Lifecycle Workflow Stages like Figure 7 As shown, the method of the present invention is divided into five stages throughout the scanner's entire lifecycle: Phase 1: Initial Calibration Phase (One-time operation after installation or major overhaul). After a 30-minute warm-up, use a standard color chart to establish full color gamut calibration. Subsequently, 50 to 100 slices were scanned to establish background, dark current, and control chart parameters, and the color cards were collected and stored in the verification library.

[0091] Phase Two: Daily Automatic Monitoring Phase (Continuous Operation). Dark field frames are automatically acquired before each batch of scans. During scanning, images are captured in real time to extract background chromaticity. Pixel differences in tile overlap areas are calculated in parallel during tile stitching. SPC control charts are monitored in real time, with zero manual intervention throughout the entire process.

[0092] Phase 3: Automatic Correction Phase (Triggered on Demand). When an interpretable abnormal state is determined, the system automatically estimates the gain change, generates correction parameters through exponential prediction and Kalman filtering, applies compensation, and updates the working baseline after successful verification.

[0093] Phase Four: Unexplained Anomaly Verification Phase (Extremely Rare Trigger). When an unexplained anomaly is detected, the system issues an alarm and suspends automatic correction, prompting the operator to insert a standard color chart for full color gamut verification. After verification, the baseline is updated and a new round of monitoring begins.

[0094] Phase 5: Anomaly Handling Phase (Rare Trigger). If the color difference still exceeds the anomaly threshold after verification with the standard color chart, manual inspection will be carried out (replacing the light source, cleaning the optical components, inspecting the sensor, etc.). After the inspection is completed, Phase 1 will be executed again.

[0095] Example 10: System hardware and software implementation, electronic devices, and readable storage media The system of this invention relies on the standard interface of the scanner host at the hardware interface level: a software-controllable shutter mechanism or light shield, readable sensor temperature data, readable and writable exposure time and channel gain registers, and a real-time WSI image data access interface. The recommended configuration for the control computer is as follows: Multi-core CPU (8 cores or more, clock speed 3.0GHz or more), 16GB or more of memory, 500GB or more of SSD storage, gigabit Ethernet interface. Computationally intensive modules can utilize GPU acceleration to keep the time for single-batch 3D detection within 10 seconds without affecting normal scanning throughput.

[0096] The software implementation includes: a dark field frame acquisition and analysis module, a full-slice digital image background region extraction and statistics module, a tile overlap area registration and verification module, a transfer function decomposition and component monitoring module, an anomaly classification decision module, a prediction model and Kalman filter module, a correction matrix calculation and application module, a statistical process control chart module, a correction log and rollback module, and a human-computer interaction interface module.

[0097] The present invention also provides 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 the above-described method embodiments. The present invention also provides a computer-readable storage medium storing computer program code thereon, which, when executed by a processor, implements the above-described method embodiments.

[0098] Example 11: Comparative Verification Experiment of New and Old Light Source Scanners To verify the effectiveness of this invention, a comparative experiment was designed between an old LED light source (with a cumulative operating time of approximately 8000 hours, referred to as the old unit) and a brand-new LED light source (with a cumulative operating time of less than 100 hours, referred to as the new unit): Experiment 1: Comparison of LED light source emission spectra. The emission spectra of the old and new LEDs were measured using a spectroradiometer (380 to 780 nm, 1 nm increment). It was expected that the peak intensity of yellow light in the old LED would decrease by 12% to 20% (phosphor degradation), the blue-yellow light power ratio would increase by 8% to 15%, and the correlated color temperature would increase by 200 to 500 K, confirming the physical mechanism by which phosphor degradation leads to an increase in color temperature.

[0099] Experiment 2: Full Color Gamut Detection and Comparison of Standard Color Chart. The old and new machines scanned the same IT8.7 / 1 color chart respectively, and the CIEDE2000 color difference was calculated. The average expected new model The average price of old machines The range is from 4.0 to 7.0, with a maximum of The red color block shifted the most when the value reached 8.0 to 15.0, proving that the old machine had a significant color deviation.

[0100] Experiment 3: Zero-reference comparison of dark-field frames. Dark-field frames were acquired with the shutter closed. It was expected that the dark current of the three channels of the old camera would increase by 10% to 35% and the spatial variance would increase by 20% to 50% compared with the new camera, confirming that dark-field frames can distinguish sensor aging.

[0101] Experiment 4: Comparison of Background Region Chromaticity Ensemble Monitoring. Using 50 blank slides from the same batch for scanning, the old machine was expected to show a 3% to 7% decrease in the red channel and a 2% to 5% increase in the blue channel in the background. Chromaticity coordinates... The gain coefficient decreased by 0.005 to 0.015. It conforms to the degradation characteristics of LED phosphors; new machine There is no obvious color cast.

[0102] Experiment 5: Comparison of tile overlap area consistency. (Expected old machine) The mean and variance were significantly higher than those of the new machine, confirming the degradation of spatial homogeneity.

[0103] Experiment 6: Continuous operation and color shift monitoring comparison. The new and old machines ran continuously for 8 hours, with sampling every 30 minutes. It was expected that the color temperature shift of the new machine would be less than 50K and the gain coefficient would remain basically unchanged; the initial color temperature of the old machine was already too high and the drift rate was faster, with the cumulative drift after 8 hours reaching 2 to 3 times that of the new machine.

[0104] Experiment 7: Comparison and Verification of Rapid Calibration Effects. The old machine was fitted with the self-reference calibration method of this invention, without inserting a standard color chart throughout the process. The expected average calibration result for the old machine was... The accuracy decreased from 4.0 to 7.0 to 1.5 to 3.0 (an improvement of 50% to 65%), and the background chromaticity shift decreased from 0.005 to 0.015 to less than 0.003, approaching the level of a new device.

[0105] Experiment 8: Verification of the self-reference system's anomaly identification capability. The experimental data above was input into the self-reference detection system. The system was expected to correctly identify the new machine as being in normal operating condition and the old machine as an explainable anomaly requiring rapid correction without triggering the standard color chart fallback verification, demonstrating that the three-dimensional self-reference monitoring can effectively distinguish different degradation states.

[0106] Example 12: Multi-scanner collaborative management extension and explanation of unwritten content The predictive model and correction parameters of this invention can be extended to collaborative management scenarios involving multiple scanners. Monitoring data from multiple scanners deployed at different sites are aggregated into a unified data platform. Using the color benchmark of one scanner calibrated with a standard color chart as a reference, the deviations of the remaining scanners from their respective three-dimensional self-reference detection systems are converted to a unified benchmark. This achieves color consistency management among multiple scanners, providing a unified color benchmark for multi-center pathological research and remote consultations. Simultaneously, time-series data from each dimension are continuously recorded to form a complete device color health record. Through population statistical learning, a more accurate degradation model is established, continuously optimizing prediction accuracy and assisting in device health maintenance.

[0107] Regarding the content not included: The technical solution of this invention is based on a statistical premise that the optical characteristics of the batches of slides used in routine scanning remain stable in a large-sample ensemble statistical manner. Background baseline data can be updated by re-collecting batch background statistics when the slide batch changes. The specific detection and baseline reconstruction process for batch changes is not part of the core subject matter claimed in this invention, and therefore will not be described further here. Furthermore, the experimental data in Example 11 are expected verification results derived from physical mechanisms, and those skilled in the art can fully implement verification based on the experimental conditions disclosed above.

[0108] 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.

[0109] The above embodiments are merely illustrative of several implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. An automated detection and correction method for color changes in a digital pathology scanner, characterized in that, Includes the following steps: During scanner installation or after major overhaul, use a standard color chart to perform baseline color calibration and obtain initial color characterization parameters and baseline reference data. During the daily operation of the scanner, a self-reference detection system is constructed, which includes a dark field frame zero reference dimension, a background chromaticity ensemble reference dimension, and a spatial redundancy check dimension. Each dimension is independent of the others and is redundant to each other. Based on the self-reference signal acquired in real time by the self-reference detection system, the drift state of different physical components in the color transfer function of the detection scanner is decoupled and the detected drift state is divided into normal state, explainable abnormal state and unexplainable abnormal state according to the preset rules. In response to whether the detected drift state is a normal state or an interpretable abnormal state, correction parameters are dynamically generated based on the degradation prediction model and the measured value of the self-reference signal, and color compensation is applied to the scanner. In response to the detected drift state being an unexplained anomaly, automatic correction is paused and full color gamut verification of the standard color chart is triggered.

2. The automated detection and correction method for color changes in a digital pathology scanner as described in claim 1, characterized in that, The zero-reference dimension of the dark field frame includes: Before scanning each batch of slices, the light path is blocked by at least one of the following methods: turning off the light source, closing the shutter, or blocking the light path. Multiple dark field images are acquired and the average value is calculated. The average pixel value of the dark field image in each color channel is calculated to obtain the current dark current. Based on the real-time operating temperature of the sensor, a preset dark current temperature index model is input to verify the conformity of the current dark current with temperature drift, so as to distinguish between normal temperature drift that conforms to the temperature model and dark current anomalies that exceed the model prediction. During image generation, baseline correction is performed by subtracting the current dark current of the corresponding channel from the original count value of each pixel.

3. The automated detection and correction method for color changes in a digital pathology scanner as described in claim 1, characterized in that, The background chromaticity ensemble reference dimensions include: Foreground and background segmentation is performed in the low-resolution pyramid layer of the sliced ​​panoramic image, and a tissue binary mask is generated and upsampled to the analysis resolution layer. Based on the area threshold, the distance constraint from the tissue boundary, and the variance of color distribution within the region, effective background candidate regions are selected. Pixel sampling is performed within the effective background candidate region, and the truncated mean is calculated to obtain the background representative color of the slice. Ensemble statistical modeling is performed on the background representative colors of multiple slices within a single batch. Normalized chromaticity coordinates and channel mean values ​​are calculated. The fluctuation of the batch background mean value converges as the sample size increases, serving as a stable virtual reference. The type of spectral drift of the light source can be identified by the direction and magnitude of the offset of the normalized chromaticity coordinates relative to the reference state.

4. The automated detection and correction method for color changes in a digital pathology scanner as described in claim 1, characterized in that, The drift states of different physical components in the color transfer function of a decoupled detection scanner include: The color transfer function is decomposed into dark current component, light source and optical transmittance combined spectral component, electronic gain component, and spatial uniformity component. Among them, the dark current component is monitored by the zero reference dimension of the dark field frame, the joint spectral component is monitored by the background chromaticity ensemble reference dimension, the electronic gain component is monitored by the joint analysis of the dark field frame and the background chromaticity, and the spatial uniformity component is monitored by the spatial redundancy check dimension. In the spatial redundancy check dimension, the pixel pairs of adjacent tiles at the same physical location in the overlapping area are extracted according to the scanning tile layout parameters, the pixel difference statistics of each color channel are calculated, and orthogonal cross-validation is performed with the detection results of the dark field frame zero reference dimension and the background chromaticity ensemble reference dimension. When an anomaly is detected in only a single dimension, the system issues an alert and continues monitoring; when anomalies occur in at least two dimensions simultaneously, it determines that the color transfer function has undergone systematic degradation and initiates a correction decision.

5. The automated detection and correction method for color changes in a digital pathology scanner as described in claim 1, characterized in that, Based on the degradation prediction model and the measured values ​​of the self-reference signal, correction parameters are dynamically generated, and color compensation is applied to the scanner, including: After deducting the effects of the current dark current and the initial reference dark current, calculate the measured values ​​of the channel gain change for each color channel; The gain prediction value is obtained based on the exponential attenuation prediction model of light source attenuation, and the gain prediction value is dynamically fused with the channel gain change measurement value through the Kalman filter algorithm to obtain the fused gain coefficient of each channel. A color correction matrix is ​​constructed based on the fusion gain coefficient to perform digital correction on the scanned image data, or the gain adjustment parameters obtained based on the fusion gain coefficient are directly written into the scanner's hardware register.

6. The automated detection and correction method for color changes in a digital pathology scanner as described in claim 1, characterized in that, Based on preset rules, the detected drift states are classified into normal states, explainable anomalous states, and unexplainable anomalous states, including: The batch self-reference statistical sequence is monitored using statistical process control methods, which include Shewhart control charts for monitoring sudden large shifts, cumulative sum control charts for monitoring small and persistent drifts, and exponentially weighted moving average control charts for monitoring mean smoothing trends. When the drift is within the confidence interval of the degradation prediction model and all control charts are in a controlled state, it is judged as a normal state; When the drift exceeds the controlled threshold, but the drift direction and evolution rate are consistent with the known physical degradation model, it is determined to be an interpretable anomalous state. When the drift amount exhibits a non-temperature-related step change, or when the drift direction is orthogonal to the historical evolution trajectory, it is determined to be an unexplainable abnormal state.

7. The automated detection and correction method for color changes in a digital pathology scanner as described in claim 1, characterized in that, After applying color compensation to the scanner, the process also includes calibration parameter verification and rollback steps: Obtain the color data of the reference slice in the compensated image, and monitor the background chromaticity convergence status of subsequent scan batches after applying compensation; The calibration verification is deemed to have failed when the color deviation of the reference slice exceeds the preset tolerance range or the background chromaticity of subsequent scan batches does not return to the reference range. In response to a calibration verification failure, the system automatically rolls back to the previous set of valid calibration parameters, records the calibration log, and triggers full color gamut verification of the standard color chart.

8. An automated detection and correction device for color changes in a digital pathology scanner, characterized in that, include: The initial calibration module is used to perform baseline color calibration using a standard color chart during scanner installation or after major repair, and to obtain initial color characteristic parameters and baseline reference data. The self-reference detection module is used to build a self-reference detection system during the daily operation of the scanner, which includes a dark field frame zero reference unit, a background chromaticity ensemble reference unit, and a spatial redundancy check unit. Each unit is independent of the others and is redundant to each other. An anomaly classification module is used to decouple the drift state of different physical components in the color transfer function of the scanner based on the self-reference signal acquired in real time by the self-reference detection system, and classify the detected drift state into normal state, explainable anomaly state and unexplainable anomaly state according to preset rules. An automatic correction module is used to dynamically generate correction parameters based on the degradation prediction model and the measured values ​​of the self-reference signal in response to whether the detected drift state is a normal state or an interpretable abnormal state, and to apply color compensation to the scanner. The fallback trigger module is used to pause automatic correction and trigger full color gamut verification of the standard color chart in response to the detected drift state being an unexplainable abnormal state.

9. 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 the automated detection and correction method for color changes in a digital pathology scanner as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the automated detection and correction method for color changes in a digital pathology scanner as described in any one of claims 1 to 7.

Images