Method and device for identifying interface hepatitis and evaluating inflammatory activity degree of liver biopsy based on pathological structure constraint, and medium

Abstract

The present application provides a liver biopsy interface inflammation recognition and inflammation activity evaluation method and device based on pathological structure constraint, and medium. The method acquires a staining whole section image of liver biopsy, extracts an effective tissue area and divides it into image blocks, and records spatial coordinates. An inflammation recognition model is used to generate an inflammation probability map and a pathological structure candidate map containing candidate areas of portal tract, fibrous septum and interface area. An interface inflammation spatial evidence map is generated according to the spatial correlation between the inflammation positive area and the candidate area, from which whole section level features are extracted, and the recognition result of interface inflammation related inflammation infiltration and the inflammation activity evaluation result are output. The present application can distinguish between intralobular scattered inflammation infiltration and interface inflammation related inflammation infiltration, and improve the interpretability and review efficiency of the evaluation result.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and computational pathology, and in particular to a method, device, and medium for identifying and assessing inflammatory activity at the interface of liver biopsy based on pathological structural constraints. Background Technology

[0002] Liver biopsy is an important tool for histopathological evaluation of chronic liver disease. When assessing the inflammatory activity of liver tissue, pathologists need to observe not only the number of inflammatory cells but also determine the anatomical location of the inflammatory infiltration and its spatial relationship with the portal area, fibrous septa, and adjacent liver parenchyma. Inflammatory infiltration located in the portal area or adjacent interface areas of the fibrous septa (i.e., interface inflammation) has a different pathological significance than scattered inflammatory infiltration within the lobules; the former is a key indicator for assessing the inflammatory activity of chronic hepatitis.

[0003] Existing digital pathology analysis methods can typically detect inflammatory areas, inflammatory cells, or inflammatory-positive image patches through image processing techniques, and can also statistically analyze quantitative indicators such as the number of inflammatory foci, the area of ​​inflammatory regions, or the positive rate. However, these methods usually only average, count, or summarize the local detection results, without considering the spatial relationship between inflammatory infiltration and pathological structures such as portal canals and fibrous septa. This leads to two typical biases: on the one hand, scattered inflammatory infiltrations within lobules far from portal canals and fibrous septa may be over-included in the assessment of inflammatory activity; on the other hand, interface-related inflammatory infiltrations that are confined to structural junctions and have a small area may be underestimated. In addition, liver biopsy samples are usually long and thin strips or multiple tissue fragments, and there are significant differences between different cases in terms of effective tissue area, tissue sampling length, number of portal canals, distribution of fibrous septa, staining quality, and scanning quality. Existing methods lack corresponding quality control mechanisms and spatial location preservation mechanisms, resulting in low interpretability and review efficiency of the assessment results.

[0004] Therefore, there is an urgent need for a method, device, and medium for identifying liver biopsy interface inflammation and assessing inflammatory activity based on pathological structural constraints, in order to solve the problems existing in the current technology. Summary of the Invention

[0005] This invention provides a method, device, and medium for identifying interface inflammation and assessing inflammatory activity in liver biopsy based on pathological structural constraints. It addresses the problems of existing methods that only summarize the whole slice based on the number of inflammatory foci, inflammation density, or positive rate, which cannot distinguish between scattered intralobular inflammatory infiltration and interface inflammation-related inflammatory infiltration. Furthermore, the assessment results lack spatial traceability, interpretability, and have low verification efficiency.

[0006] The core technology of this invention is to simultaneously generate an inflammation probability map and a pathological structure candidate map containing candidate regions of the portal region, fibrous septum, and interface region in the same whole-slice image coordinate system. Based on spatial correlation features such as distance distribution, neighborhood enrichment, adjacency relationship, and continuous distribution between the inflammatory positive region and the pathological structure candidate region, a spatial evidence map of interface inflammation is generated. By summarizing the structural instance-level features, the identification results of interface inflammation-related inflammatory infiltration and the inflammatory activity assessment results are output, thereby realizing the distinction between interface inflammation-related inflammatory infiltration and scattered inflammatory infiltration within the lobules.

[0007] In a first aspect, the present invention provides a method for identifying and assessing inflammatory activity at the interface of liver biopsy based on pathological structural constraints, the method comprising the following steps:

[0008] Obtain stained whole-section images of liver biopsy; Extract the effective tissue region from the whole slice image, divide the effective tissue region into multiple image blocks, and record the spatial coordinates of the effective tissue region and image blocks in the coordinate system of the whole slice image. An inflammation probability map is generated by using an inflammation recognition model to predict the inflammation probability of image patches or pixel regions mapped from image patches. Generate a candidate map of pathological structures in the same coordinate system as the inflammation probability map. The candidate map of pathological structures includes a candidate region of the portal area, a candidate region of the fibrous septum, and a candidate region of the interface area. The candidate region of the interface area is obtained by extending from the boundary of the portal area or the boundary of the fibrous septum towards the adjacent liver parenchyma. Based on the spatial correlation between the inflammatory positive areas and the candidate areas of the portal canal, fibrous septum, or interface region, a spatial evidence map of interface inflammation is generated. Inflammation positive areas that are far from the candidate areas of the portal canal, fibrous septum, and interface region and do not satisfy the spatial correlation are identified as scattered inflammatory infiltration within the lobules, and the contribution of such inflammatory positive areas to the spatial evidence map of interface inflammation is reduced. Extracting full-slice-level features from spatial evidence maps of interface inflammation; and Based on the full-slice-level features, the results of identifying interface inflammation-related inflammatory infiltration and assessing inflammatory activity are output.

[0009] Furthermore, the extraction of effective tissue regions includes using threshold segmentation, color space segmentation, tissue mask extraction, morphological processing, artifact recognition, or a combination thereof, to exclude background, blank, folded, contaminated, severely blurred, or abnormally stained regions; the inflammation probability map, effective tissue region, image block coordinates, and pathological structure candidate images are unified to the same magnification or the same physical scale coordinate system. When there is a coordinate mapping between different magnifications, coordinate conversion is performed based on the pyramid level, downsampling factor, image block coordinates, and image block size of the whole slice image.

[0010] Furthermore, the spatial correlation includes at least one of the following: the distance distribution from the inflammatory positive region to the boundary of the gate region or the fiber septum boundary; the enrichment degree of the inflammatory positive region in the neighborhood of the structural boundary; the degree of overlap or adjacency between the inflammatory positive region and the candidate region of the interface region; the degree of continuous distribution of the inflammatory positive region along the structural boundary; and the proportion of affected length along the direction of the structural boundary.

[0011] Furthermore, the spatial evidence map of interface inflammation is obtained by fusing the inflammation probability map with spatial association weights; the spatial association weights include at least one of distance weights, structural neighborhood weights, and continuity weights. The distance weights reflect the distance from the location to the gate region or fiber septum boundary, the structural neighborhood weights reflect whether the location is located within the candidate region of the interface region or the neighborhood of the structural boundary, and the continuity weights reflect the degree of continuous distribution of the inflammatory positive region along the structural boundary.

[0012] Furthermore, based on the structural instances of the candidate portal region or the candidate fibrous septum region, the inflammatory load, the proportion of affected length, the maximum local evidence strength, and the degree of continuous distribution within the neighborhood of each structural instance are calculated, and full-slice-level features are obtained through maximum value, mean, quantile, weighted summation, or attention-weighted methods.

[0013] Furthermore, the assessment results of inflammatory activity include no obvious interface inflammation-related inflammatory infiltration, presence of interface inflammation-related inflammatory infiltration, mild activity, moderate activity, severe activity, continuous score, or at least one of the above results, and include the corresponding confidence information.

[0014] Furthermore, it also includes a quality control step; the quality control step generates a quality control result based on the effective tissue area, tissue sampling length, number of effective gate regions, image clarity, degree of staining abnormality, proportion of folded areas, proportion of blank areas or proportion of artifact areas; when the quality control result does not meet the preset conditions, a low confidence prompt or a manual review prompt is output.

[0015] Secondly, the present invention provides a device for identifying and assessing inflammatory activity at the liver biopsy interface based on pathological structural constraints, comprising: The image acquisition module is used to acquire stained whole-section images of liver biopsy. The preprocessing module is used to extract the effective tissue region from the full slice image, divide the effective tissue region into multiple image blocks, and record the spatial coordinates of the effective tissue region and image blocks in the coordinate system of the full slice image; The inflammation probability prediction module is used to predict the inflammation probability of image blocks or pixel regions mapped from image blocks using an inflammation recognition model, and generate an inflammation probability map. The pathological structure candidate map generation module is used to generate pathological structure candidate maps in the same coordinate system as the inflammation probability map. The pathological structure candidate map includes the portal zone candidate region, the fibrous septum candidate region, and the interface zone candidate region. The interface zone candidate region is obtained by extending from the portal zone boundary or the fibrous septum boundary towards the adjacent liver parenchyma. The spatial evidence generation module is used to generate a spatial evidence map of interface inflammation based on the spatial correlation between the inflammatory positive area and the candidate areas of the portal canal, fibrous septum, or interface region. Inflammation positive areas that are far from the candidate areas of the portal canal, fibrous septum, and interface region and do not satisfy the spatial correlation are identified as scattered inflammatory infiltration within the lobules, and the contribution of the inflammatory positive area to the spatial evidence map of interface inflammation is reduced. The evaluation output module extracts full-slice-level features from the spatial evidence map of interface inflammation and outputs the identification results of interface inflammation-related inflammatory infiltration and the assessment results of inflammatory activity based on the full-slice-level features; and The visualization verification module is used to display the inflammation probability heatmap, pathological structure candidate map, interface inflammation spatial evidence map, and inflammation activity assessment results.

[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 perform the above-described method for identifying and assessing the inflammatory activity of liver biopsy interface inflammation based on pathological structural constraints.

[0017] Fourthly, the present invention provides a readable storage medium storing a computer program, the computer program including program code for controlling a process to execute the process, the process including the above-described method for identifying and assessing inflammatory activity at the liver biopsy interface based on pathological structural constraints.

[0018] The main contributions and innovations of this invention are as follows: 1. This invention does not only count the number or area of ​​inflammatory regions, but also incorporates the spatial relationship between the probability of inflammation and the candidate regions of the portal area, fibrous septum and interface area into the analysis framework. By constructing a spatial evidence map of interface inflammation through multi-dimensional spatial correlation features such as distance distribution, neighborhood enrichment, boundary adjacency, continuous distribution and affected length, it helps to distinguish between scattered inflammatory infiltrations within the lobules and inflammatory infiltrations located in the portal area or the interface area adjacent to the fibrous septum, thereby improving the accuracy of inflammatory activity assessment.

[0019] 2. This invention preserves the spatial correspondence between the inflammatory region, the candidate region of pathological structure and the original image in the coordinate system of the whole slice image. The interface inflammatory spatial evidence map can show the main evidence region and the affected structural region to the pathologist and support jumping from the global view to the local review view, thereby improving the interpretability of the evaluation results and the review efficiency of the pathologist.

[0020] 3. This invention introduces a quality control mechanism based on multi-dimensional indicators such as effective tissue area, tissue sampling length, number of effective gate areas, image clarity, degree of staining abnormality, and artifact ratio. When the above indicators do not meet the preset conditions, a low confidence prompt or a manual review prompt can be output, which helps to avoid the undue influence of low-quality samples on the evaluation results and improves the consistency and reliability of the system among different cases.

[0021] 4. The inflammation recognition model and pathological structure candidate image generation model of the present invention do not depend on a specific network architecture. They can adopt convolutional neural networks, visual Transformers, multi-scale image patch models, weakly supervised multi-instance learning models or traditional machine learning models, which facilitates flexible deployment and implementation in different digital pathology platforms and computing environments.

[0022] 5. This invention supports pathologists in providing feedback and corrections to visualization results. The correction results recorded by the system can be used as case review records, threshold calibration samples, or subsequent model training samples, forming a closed-loop mechanism of analysis-review-improvement, which helps to continuously improve the evaluation performance of the system.

[0023] 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

[0024] 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 illustrating the process of preprocessing liver tissue images, extracting image block-level features, and summarizing full-slice-level results in one embodiment of the present invention.

[0025] Figure 2 This is an example of a liver tissue pathological image block used for training an image block recognition model in one embodiment of the present invention.

[0026] Figure 3 This is a visualization example of a whole slice image of liver tissue and analysis results under a global view in one embodiment of the present invention.

[0027] Figure 4This is a visualization example of the liver tissue region and analysis results under a partial view in one embodiment of the present invention.

[0028] Figure 5 This is a schematic diagram of the verification interface for the analysis results of a full-slice image in one embodiment of the present invention.

[0029] Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0030] 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.

[0031] 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.

[0032] Example 1 like Figure 1 As shown, this embodiment provides a method for identifying interface inflammation and assessing inflammatory activity in liver biopsy based on pathological structural constraints. The overall process of this method includes: input image acquisition and data definition, effective tissue extraction and image block segmentation, generation of inflammation probability map, generation of pathological structure candidate map, generation of interface inflammation spatial evidence map, summarization of structural instance-level features and output of results, quality control, interpretability display and pathologist review.

[0033] I. Input Image and Data Definition (S1) Step S1: Obtain stained whole-section images of liver biopsy. The input for this embodiment is one or more hematoxylin and eosin (H&E) stained whole-section images of liver biopsy. The whole-section images are obtained using a digital slide scanner, typically with a multi-layered pyramid structure, allowing for the reading of local image regions at different magnifications. Lower magnifications (e.g., 2x or 5x) are used to identify tissue distribution and structural context, while higher magnifications (e.g., 20x or 40x) are used to observe cytological details.

[0034] At the target magnification, the coordinates of the full-slice image are defined as follows: Effective tissue mask is denoted as The inflammation probability graph is denoted as Candidate areas for gate management are recorded as Candidate regions for fiber spacers are denoted as The candidate area of ​​the interface is denoted as The above figures can be binary images or probability images with confidence values, all of which are located in the same full-slice image coordinate system.

[0035] In this embodiment, the "interface candidate region" refers to the analysis area at the junction of the portal region or fibrous septum and the adjacent liver parenchyma. This region does not require a pathologically accurate depiction at every pixel level; as long as it can provide candidate locations for the spatial relationship analysis of inflammatory infiltration and related pathological structures, it can be used for subsequent calculations.

[0036] In optional implementations, Masson staining images, immunohistochemical images, or other images that can reflect liver tissue structure information can also be combined. If the same case has both H&E staining images and Masson staining images or immunohistochemical images, rigid or non-rigid registration can be completed first using tissue contours, portal areas, blood vessels, or other feature points. Then, structural information is used as supplementary input and jointly judged with the inflammation probability map of the H&E staining image in the same coordinate system. This multimodal fusion is an optional technical means and does not affect the implementation of the core solution of this invention.

[0037] II. Effective organization extraction and image block segmentation (S2) Step S2 extracts the effective tissue region from the full slice image, divides the effective tissue region into multiple image blocks, and records the spatial coordinates of the effective tissue region and image blocks in the coordinate system of the full slice image.

[0038] First, the full-slice image is read, and grayscale conversion, color space conversion, or color normalization is performed at the target magnification (20x in this embodiment). Initial tissue regions are obtained using thresholding segmentation, color space segmentation, tissue masking models, or morphological processing. OTSU thresholding segmentation automatically calculates the optimal threshold to separate the tissue foreground from the background; color space segmentation converts the image from RGB space to HSV or LAB space, and uses chroma, saturation, or luminance channels for tissue region identification; the tissue masking model can use a pre-trained semantic segmentation network to output a tissue foreground mask in an end-to-end manner.

[0039] After obtaining the initial tissue region, opening operations are used to remove fine noise, closing operations are used to bridge small breaks at tissue edges, pores are filled to seal the blank areas within the tissue, and small connected regions are used to remove debris areas with areas below a preset threshold. Areas with scan boundaries, blank backgrounds, folds, contamination, severe defocusing, or abnormal staining are excluded using color rules (such as abnormal color cast detection), sharpness indices (such as Laplacian variance), or quality control models. The effective tissue region obtained after these processes is denoted as... .

[0040] The effective tissue region is divided into image blocks. In this embodiment, it is used at 20x magnification. Image patches of pixels. The size, magnification, and step size of the image patches can be adjusted according to the scan resolution and model input requirements. Each image patch records its top-left corner coordinates, center coordinates, size, pyramid level, magnification information, and tissue percentage. Image patches with a tissue percentage below a preset threshold (e.g., below 50%) are not included in subsequent analysis.

[0041] If there is overlap between adjacent image patches, the model output of the overlapping areas can be fused by averaging, maximizing, or weighting the distance between the center of the image patches. In addition to the image patches themselves, the preprocessing also records information such as brightness, color mean, color standard deviation, sharpness, tissue proportion, and artifact markers. This information can be used to filter low-quality image patches before inference, or to generate case-level quality control results.

[0042] When analysis is required at different magnifications, the inflammation probability map, effective tissue region, image block coordinates, and candidate pathological structures are unified to the same magnification or the same physical scale coordinate system. Specifically, coordinate conversion is performed based on the pyramid level of the whole-slice image, the downsampling factor, the coordinates of the top-left corner of the image block, and the image block size. Image data at different magnifications are first converted to the same physical coordinates according to the whole-slice pyramid level, physical pixel size, and the coordinates of the top-left corner of the image block. Before fusion, registration deviations are checked using tissue contours, structural boundaries, or manual markers; if the deviation exceeds the allowable range, re-registration is performed or manual verification is prompted.

[0043] III. Generation of Inflammation Probability Map (S3) Step S3 uses an inflammation recognition model to predict the inflammation probability of image blocks or pixel regions mapped from image blocks, generating an inflammation probability map.

[0044] Inflammation recognition models can employ convolutional neural networks, visual Transformers, multi-scale image patch models, weakly supervised multi-instance learning models, traditional machine learning models, or combinations thereof.

[0045] In one implementation, the inflammation recognition model can use ResNet-50 as the backbone network. The RGB image patch of the pixel is used as input. After feature extraction through multiple convolutional and pooling layers, the inflammation probability value between 0 and 1 is output through a fully connected layer and a softmax layer.

[0046] In another implementation, a visual Transformer (such as ViT-B / 16) can be used to divide the image patch into multiple The pixel patches are analyzed using a multi-head attention mechanism to calculate the correlation between each patch, and then the inflammation probability is predicted after fusing global information.

[0047] In another implementation, a weakly supervised multi-instance learning model can be used, with the whole slice image as the bag and the image patch as the instance. The model is trained using the whole slice level labels, and the network automatically learns the contribution weight of each image patch to the final prediction result.

[0048] In another implementation, a multi-scale model combining low-magnification context and high-magnification detail can be used, with low-magnification image patches (e.g., 5x, ...) Pixels provide organizational context, high-magnification image patches (such as 20x, (pixels) provide cellular details, and the two are fused together to output the probability of inflammation.

[0049] The inflammation recognition model can be trained under supervision by positive and negative inflammation image patches labeled by pathologists, or it can be trained using a weakly supervised multi-instance learning approach, which uses full-slice-level labels to constrain the model.

[0050] Based on the original coordinates of the image patch, the inflammation probability value output by the model is backfilled into the coordinate system of the whole slice image to obtain the inflammation probability map. The inflammation probability map can be an image block-level probability map (each image block corresponds to a probability value), a pixel-level probability map (the probability is output pixel by pixel by the semantic segmentation model), or a probability heatmap obtained through coordinate mapping and interpolation. When the output is an image block-level probability, the probability value is mapped to the pixel position within the image block through bilinear interpolation or nearest neighbor interpolation to form a pixel-level probability heatmap.

[0051] exist Figure 1In the tile-level training framework shown, the model training employs a multi-scale strategy. Image tiles at 10x and 20x magnification are extracted from the full-slice image, respectively, and after data augmentation, are input into the model for feature extraction. During training, classification loss and similarity loss between features at different magnifications are calculated simultaneously to constrain the consistency of feature representation at the same location under different magnifications, thereby improving the model's ability to recognize cross-scale inflammatory features. Furthermore, in the WSI-level inference framework, the features of the image tiles and their location embedding information are aggregated and used together for the final classification. The location embedding is used to preserve the spatial location information of the image tiles in the full-slice image.

[0052] IV. Generation of Candidate Pathological Structure Images (S4) Step S4 is related to the inflammation probability map. Candidate maps of pathological structures are generated within the same coordinate system as the whole-slice image. These candidate maps include portal region candidate areas. Candidate regions for fiber spacers and interface candidate area The regions mentioned above can be binary maps or probability maps with confidence values.

[0053] Candidate images of pathological structures can be generated by image segmentation models (such as U-Net), pathological region detection models (such as Faster R-CNN), color and texture rules, pathologist-annotated training models or weakly annotated training models, or combinations thereof.

[0054] Candidate regions for the portal canal can be identified based on the morphology of small bile ducts, blood vessel lumens, connective tissue, and surrounding cells. Candidate regions for fibrous septa can be identified based on the color and texture characteristics of the fibrous tissue septa.

[0055] Candidate area in the interface There are two methods for generating the interface region. The first method involves obtaining the portal zone boundary and fibrous septum boundary, then extending these boundaries towards the adjacent liver parenchyma within a predetermined range to obtain candidate regions for the interface. The extension range can be determined by combining scan resolution, structure size, and pathologist annotations. The second method involves directly outputting the candidate interface region from the structure recognition model.

[0056] If structural boundaries are difficult to segment accurately and stably, a structural heatmap or high-confidence candidate region can be obtained first, and then the analysis neighborhood can be constructed through dilation, erosion, skeletonization, boundary extraction, or distance transformation. For structural candidate regions near the outer edge of tissue strips, cracks, cavities, folds, and broken areas, they can be marked as exclusion regions or low-weight regions to reduce interference caused by artifacts.

[0057] During training, at least two pathologists should mark or review the portal area, fibrous septa, and interface area. If there is a disagreement, they should review the area together or have it confirmed by a third pathologist.

[0058] V. Generation of Spatial Evidence Map of Interface Inflammation (S5) Step S5 will generate an inflammation probability map. After spatially mapping the candidate pathological structures, a spatial evidence map of interface inflammation is generated based on the spatial correlation between the inflammatory positive regions and each candidate region. This spatial evidence map is not simply an inflammation heatmap, but rather an intermediate result recording the relationship between the probability of inflammation and the location of pathological structures.

[0059] Spatial relationships include at least one of the following, which can be divided into two categories: spatial metric features and boundary features.

[0060] The first category is spatial metric features, including at least one of the following: First, the distribution of distances from inflammatory positive areas to the boundaries of the portal canal or fibrous septa. This feature reflects the spatial distance between the inflammatory positive area and the nearest portal canal or fibrous septum boundary; the closer the distance, the higher the likelihood of it being evidence of interface inflammation.

[0061] Second, the enrichment degree of inflammatory positive areas within the neighborhood of the structural boundary. This enrichment degree refers to the ratio of the density of inflammatory positive pixels within the neighborhood of the structural boundary to the average density of inflammatory pixels in the effective tissue area of ​​the same whole slice. The higher the ratio, the higher the concentration of inflammation near the structural boundary, and the stronger the evidence related to interface inflammation.

[0062] Third, the degree of overlap or adjacency between the inflammatory positive region and the interface candidate region. The degree of overlap refers to the ratio of the intersection area to the union area of ​​the inflammatory positive region and the interface candidate region at the pixel level; the degree of adjacency refers to the ratio of the contact length between the two at the edge to the perimeter of the interface candidate region.

[0063] The second category is features along the boundary, including at least one of the following: Fourth, the degree of continuous distribution of inflammatory positive areas along structural boundaries. This feature assesses whether inflammatory positive areas form a continuous or near-continuous band-like distribution along the boundaries of the portal canal or fibrous septa, rather than a scattered point-like distribution.

[0064] Fifth, the proportion of affected length along the structural boundary. The denominator of this proportion is the total length of the boundary between the structural instance and the liver parenchyma that can be used for determination; the slice edge, fragmented area, and artifact area are not included in the denominator. The numerator is the length of the boundary of the structural instance affected by the inflammatory positive area.

[0065] In one specific implementation, the interface inflammation spatial evidence map is obtained by fusing the inflammation probability map with spatial association weights, and can be represented as:

[0066] in, Distance weights reflect location. Distance to the gate area or fiber spacer boundary; The structural neighborhood weights reflect location. Whether it is located within the candidate region of the interface area or the neighborhood of the structural boundary; The continuity weight reflects the degree of continuous distribution of inflammatory positive areas along the structural boundary.

[0067] The specific calculation method for distance weights is as follows: Let the set of structural boundaries be... ,Location The shortest distance to this boundary set is The distance weight is then expressed as:

[0068] in This is a preset distance attenuation parameter used to control the rate at which distance reduces the strength of evidence. The value can be determined based on the scan resolution, structure size, and pathological experience.

[0069] Structural Neighborhood Weights According to location Whether it falls within the interface region candidate area, gate region boundary buffer, or fiber spacer boundary buffer is determined. If the location... Falling into any of the above areas, Take the higher value; otherwise, take the lower value or zero.

[0070] Continuity weight The continuity weight is determined based on the contact length between the inflammatory positive connected region and the structural boundary, the adjacency relationship, and the degree of continuous coverage along the boundary direction. Continuously distributed inflammatory positive regions receive a higher continuity weight, while isolated or discontinuously distributed inflammatory positive regions receive a lower continuity weight.

[0071] In another extended implementation, the spatial evidence map can further incorporate the credibility of structural candidate regions. Boundary adjacency weight Neighborhood enrichment weight Image and registration quality weights , is represented as:

[0072] Boundary adjacency and quality control are both included as independent factors in the calculation.

[0073] For inflammatory-positive regions that are far from the candidate regions of the portal region, fibrous septum, and interface region and do not meet the above spatial correlations, the system identifies them as scattered intralobular inflammatory infiltrations and reduces their contribution to the spatial evidence map of interface inflammation. The system retains these regions as indications of intralobular inflammation but does not treat them as high-weighted evidence of interface inflammation, thus distinguishing between scattered intralobular inflammation and interface inflammation-related inflammation.

[0074] Spatial evidence maps can be single-channel or multi-channel images. Multi-channel images can record information such as inflammation probability, structural neighborhood confidence, distance weight, boundary adjacency, continuity distribution, and quality control weights.

[0075] VI. Full-slice-level feature extraction and result output (S6) Step S6 extracts full-slice-level features from the spatial evidence map of interface inflammation, and outputs the identification results of interface inflammation-related inflammatory infiltration and the assessment results of inflammatory activity based on the full-slice-level features.

[0076] The candidate regions for the gate region and the candidate regions for the fiber spacers are decomposed into multiple structural instances. For the first A structure instance, in its neighborhood The following features are calculated internally: Inflammatory burden : The weighted area or weighted probability sum of inflammatory positive regions within the neighborhood of the structural instance.

[0077] Affected length ratio The ratio of the length of the boundary of the structure instance affected by the inflammatory positive region to the total length of the boundary that can be used for determination.

[0078] Maximum local evidence strength : The maximum value of the interface inflammation spatial evidence map within the neighborhood of this structural instance.

[0079] Continuous distribution degree The degree of continuous coverage of the inflammatory positive area along the boundary direction of the structural instance.

[0080] The structural instance-level score is represented as follows:

[0081] in For quality control weights, , , and These are preset weights or weights obtained through training.

[0082] Based on the scores of multiple structural instances, full-slice-level features are generated using maximum, mean, quantile, weighted summation, or attention-weighted methods. These full-slice-level features may include the inflammatory load in the interface region neighborhood, the number of affected portal areas, the number of affected fibrous septa, the proportion of affected lengths, the intensity of the main evidence region, the intralobular inflammatory load, the effective tissue area, the tissue sampling length, and quality control indicators.

[0083] The evaluation module can employ threshold rules, logistic regression, support vector machines, random forests, gradient boosting trees, neural networks, or learnable aggregation networks to output results. In one implementation, a learnable aggregation model is used, taking as input features such as spatial evidence, affected length, neighborhood inflammation density, continuous distribution, and image quality for each structural instance, and outputting the probability of interface inflammation presence and activity score; this aggregation model can employ a multilayer perceptron or gradient boosting tree and be trained using pathologist-annotated results.

[0084] The output includes the identification results of interface inflammation-related inflammatory infiltration and the assessment results of inflammatory activity. The inflammatory activity assessment results include no obvious interface inflammation-related inflammatory infiltration, presence of interface inflammation-related inflammatory infiltration, mild activity, moderate activity, severe activity, continuous score, or at least one of the above results, and include corresponding confidence information. Activity is graded according to a pre-determined pathological scoring caliber. Continuous scores can be further mapped to none, mild, moderate, and severe; specific cutoff values ​​are determined by the pathologists' review results in the training set and are consistently used in the same version of the model.

[0085] For cases where the quality control results do not meet the requirements, the system will simultaneously output a low confidence warning and explain the possible reasons, such as insufficient effective tissue, insufficient number of effective portal areas, blurred images, or overlap between the main evidence area and the artifact area.

[0086] VII. Quality Control This method also includes a quality control step, which plays a role in multiple stages of the image analysis workflow. Quality control comprehensively evaluates at least some of the following indicators to generate quality control results: effective tissue area, tissue sampling length, number of effective gate regions, image sharpness, degree of staining abnormalities, proportion of folded regions, proportion of blank regions, and proportion of artifact regions.

[0087] When the quality control results do not meet the preset conditions, the system will not output a high-confidence conclusion, but will instead output a low-confidence prompt or a manual review prompt, explaining possible reasons, such as insufficient effective tissue, insufficient number of effective gate areas, blurry image, or overlap between the main evidence area and the artifact area. When key indicators fail to meet the standards, the system may prompt for rescanning or manual review.

[0088] Quality control weight It can participate in the structural instance-level scoring calculation, and for structural instances with low quality assessment, its contribution to the full slice-level features is reduced.

[0089] 8. Explanatory presentation and pathologist review The system displays the analysis results using a layered overlay method. See also Figure 3 The system displays the original image of the whole slice in a global view. Figure 3 Part a) and AI-visualized diagnostic images ( Figure 3 (Part b) The AI ​​diagnostic image highlights the abnormal areas identified by the AI ​​in deep purplish-red, which perfectly correspond to the tissue contours of the original image. See also Figure 4 The system displays the original image of the selected area within a local field of view. Figure 4 The image (part a) and AI-annotated images (part b) are presented with translucent reddish-brown polygonal areas covering the AI-identified abnormal areas, facilitating pathologists' observation of evidence details at high magnification. The display layers include: the first layer showing the original pathological image; the second layer showing an inflammation probability heatmap; the third layer showing boundary annotations for candidate regions of the portal, fibrous septa, and interface areas; the fourth layer showing the main spatial evidence areas; and the fifth layer showing the assessment results, confidence levels, and quality control prompts. Pathologists can switch between layers and jump from the global view to a local view of the main evidence areas to observe evidence details at high magnification.

[0090] Interpretable display results include at least one of the following: inflammation probability heatmap, pathological structure candidate map, interface inflammation spatial evidence map, main evidence region, affected structure region, and local review view.

[0091] See Figure 5 The review interface includes a navigation area, a field-of-view switching button, a zoom control bar, a layer visibility switch, and diagnostic result cards. Pathologists can use the layer visibility switch to turn on or off annotation layers such as the inflammation heatmap, structural boundaries, evidence atlas, and tissue contours. The inflammation heatmap uses a blue-to-red gradient color scheme, with blue representing low inflammation risk and red representing high inflammation risk. The diagnostic result cards on the right side of the interface display the positive or negative inflammatory assessment and confidence level, the percentage of high-risk areas on the inflammation heatmap, the percentage of structural boundary length, and thumbnails of high-confidence local evidence atlases. The physician review module provides three options: "Positive," "Negative," and "Indeterminate," as well as a remarks input box. When a pathologist believes that a certain area has been misdiagnosed or missed, they can correct the candidate structural area, the main evidence area, or the assessment result in the review interface. The system saves the original model output along with the review results, recording the modification time, modified area, and modification type. The review results can be used for case review records, threshold calibration, and subsequent model training, forming a closed-loop mechanism of analysis-review-improvement.

[0092] IX. Training and Verification During the training phase, the training process for each model is as follows: Training of the inflammation recognition model: Supervised training was performed using inflammation-positive and inflammation-negative image patches labeled by pathologists. For example... Figure 2 As shown, the training samples include inflammatory positive image blocks and inflammatory negative image blocks labeled by pathologists according to a unified legend, covering a variety of liver pathological states, including ballooning degeneration, watery degeneration, normal liver tissue, fibrosis, hepatitis B positive, hepatitis B negative, inflammatory activity positive, inflammatory activity negative, and fatty degeneration, to enhance the model's ability to identify different pathological features.

[0093] Training the structural recognition model: Training is performed using boundary or region annotations of portal regions, fiber septa, or interface regions. When structural annotations are limited, initial candidate regions can be generated using color and texture rules, and then the model can be trained using a small number of corrected samples from pathologists; alternatively, weakly supervised learning and full-slice-level labels can be used to constrain the model.

[0094] Training the inflammation activity assessment model: A summary model or a learnable aggregation network is trained using full-slice-level labels. When pathologists provide feedback and corrections to the analysis results, the system-recorded corrections can be used as threshold calibration samples or subsequent model training samples.

[0095] Regarding annotation requirements, the training set must be annotated by at least two pathologists using a unified legend. Inflammation can be annotated by pixels or image blocks; portal areas, fibrous septa, and interface areas should be annotated by boundaries or regions; if there are discrepancies, they should be jointly reviewed, and if they still cannot be consistent, a third pathologist should make the final decision.

[0096] During the validation phase, cases are divided into training, validation, and test sets. The consistency, sensitivity, specificity, area under the receiver operating characteristic (AUD) curve, or graded consistency rate between the identification results and the pathologist's review results are statistically analyzed. For multi-center data, stratified validation can be performed by scanner model, staining batch, medical institution origin, or disease subtype to assess the model's generalization ability.

[0097] Example 2 This embodiment also provides a system for identifying and assessing inflammatory activity at the liver biopsy interface based on pathological structural constraints, including the following modules.

[0098] The image acquisition module is used to acquire stained whole-section images of liver biopsies. This module connects to a digital slide scanner, receives the whole-section image data generated by the scan, and transmits the image data to the preprocessing module.

[0099] The preprocessing module extracts the effective tissue region from the whole-slice image, divides the effective tissue region into multiple image blocks, and records the spatial coordinates of the effective tissue region and image blocks in the whole-slice image coordinate system. The preprocessing module also records the coordinates, magnification, and size information of each image block in the whole-slice pyramid hierarchy, establishes the transformation relationship from multi-magnification image blocks to a unified coordinate system, and transmits the processed data to the inflammation probability prediction module and the pathological structure candidate image generation module.

[0100] The inflammation probability prediction module is used to predict the inflammation probability of image patches or pixel regions mapped from image patches using an inflammation recognition model, and generate an inflammation probability map. This module internally deploys a pre-trained inflammation recognition model, receives image patch data output from the preprocessing module, outputs the inflammation probability of each image patch or pixel region, and fills the probability values ​​back into the coordinate system of the whole-slice image based on the coordinate information to generate the inflammation probability map.

[0101] The pathological structure candidate map generation module generates pathological structure candidate maps located in the same coordinate system as the inflammation probability map. These candidate maps include portal zone candidate regions, fibrous septum candidate regions, and interface zone candidate regions. Internally, this module deploys a structure recognition model or rule engine, receives image data from the preprocessing module, and outputs the location and confidence information of each candidate region.

[0102] The spatial evidence generation module generates a spatial evidence map of interface inflammation based on the spatial relationships between inflammatory positive regions and candidate regions of the portal, fibrous septa, or interface regions. This module receives the inflammation probability map output by the inflammation probability prediction module and the pathological structure candidate map output by the pathological structure candidate map generation module. It calculates the distance weight, structural neighborhood weight, and continuity weight for each spatial location and generates the spatial evidence map of interface inflammation through weighted fusion.

[0103] The evaluation output module extracts full-slice-level features from the spatial evidence map of interface inflammation and outputs the identification results of interface inflammation-related inflammatory infiltration and the assessment results of inflammatory activity based on the full-slice-level features. This module performs feature extraction and score aggregation by structural instance and outputs the final evaluation results through a classifier or regression model.

[0104] The visualization and review module displays inflammation probability heatmaps, candidate pathological structures, interface inflammation spatial evidence maps, and inflammation activity assessment results in a layered overlay format. This module supports layered overlay display, allowing pathologists to switch between global and local views, and provides an interactive correction interface. When pathologists correct key evidence regions, structural boundaries, or assessment results, this module records the corrections and stores them in a database for threshold calibration or subsequent model training.

[0105] The modules described above can run on general-purpose computing devices, including but not limited to workstations, servers, or cloud computing platforms. Data transfer between modules is achieved through standardized interfaces, with the full-slice image coordinate system serving as a unified reference framework for data interaction between modules.

[0106] Example 3 This embodiment also provides an electronic device, see reference. Figure 6 It includes a memory 402 and a processor 401, the memory 402 storing a computer program and the processor 401 being configured to run the computer program to perform the steps in any of the above method embodiments.

[0107] 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 the present invention.

[0108] 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.

[0109] 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.

[0110] The processor 401 reads and executes computer program instructions stored in the memory 402 to implement any of the methods for identifying liver biopsy interface inflammation and assessing inflammatory activity based on pathological structural constraints in the above embodiments.

[0111] 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.

[0112] 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.

[0113] Input / output device 404 is used for inputting or outputting information. It can be a speaker, microphone, monitor, or keyboard.

[0114] Example 4 This embodiment also provides a readable storage medium storing a computer program, the computer program including program code for controlling the process to execute the process, the process including the method for identifying liver biopsy interface inflammation and assessing inflammatory activity based on pathological structural constraints according to Embodiment 1.

[0115] 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.

[0116] 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.

[0117] 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 1 Any 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.

[0118] 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.

[0119] 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. A method for identifying interface inflammation and assessing inflammatory activity in liver biopsy based on pathological structural constraints, characterized in that, Includes the following steps: Obtain stained whole-section images of liver biopsy; Extract the effective tissue region from the full slice image, divide the effective tissue region into multiple image blocks, and record the spatial coordinates of the effective tissue region and the image blocks in the coordinate system of the full slice image; An inflammation probability map is generated by using an inflammation recognition model to predict the inflammation probability of the image block or the pixel region mapped from the image block. Generate a candidate map of pathological structures in the same coordinate system as the inflammation probability map. The candidate map of pathological structures includes a candidate region of the portal canal, a candidate region of the fibrous septum, and a candidate region of the interface. The candidate region of the interface is obtained by extending the boundary of the portal canal or the boundary of the fibrous septum towards the adjacent liver parenchyma. Based on the spatial correlation between the inflammatory positive areas and the candidate areas of the portal canal, fibrous septum, or interface region, a spatial evidence map of interface inflammation is generated. Inflammation positive areas that are far from the candidate areas of the portal canal, fibrous septum, and interface region and do not satisfy the spatial correlation are identified as scattered intralobular inflammatory infiltrations, and the contribution of such inflammatory positive areas to the spatial evidence map of interface inflammation is reduced. Extract full-slice-level features from the spatial evidence map of the interface inflammation; as well as Based on the full-slice-level features, the identification results of interface inflammation-related inflammatory infiltration and the assessment results of inflammatory activity are output. The interface inflammation spatial evidence map is obtained by fusing the inflammation probability map with spatial correlation weights. The spatial correlation weights include at least one of distance weights, structural neighborhood weights, and continuity weights. The distance weights reflect the distance from the location to the gate region or fiber septum boundary. The structural neighborhood weights reflect whether the location is located within the candidate region of the interface region or the neighborhood of the structural boundary. The continuity weights reflect the degree of continuous distribution of the inflammatory positive region along the structural boundary.

2. The method as described in claim 1, characterized in that, Extracting the effective tissue region involves using threshold segmentation, color space segmentation, tissue mask extraction, morphological processing, artifact recognition, or a combination thereof to exclude background, blank, folded, contaminated, severely blurred, or abnormally stained regions. The inflammation probability map, effective tissue region, image block coordinates, and pathological structure candidate images are unified to the same magnification or the same physical scale coordinate system. When there is a coordinate mapping between different magnifications, coordinate conversion is performed based on the pyramid level, downsampling factor, image block coordinates, and image block size of the whole slice image.

3. The method as described in claim 1, characterized in that, The spatial correlation includes at least one of the following: the distance distribution from the inflammatory positive region to the boundary of the gate region or the boundary of the fiber septum; the enrichment degree of the inflammatory positive region in the neighborhood of the structural boundary; the degree of overlap or adjacency between the inflammatory positive region and the candidate region of the interface region. The degree of continuous distribution of inflammatory positive areas along the structural boundary.

4. The method as described in claim 1, characterized in that, Based on the structural instances of candidate regions of the portal canal or fibrous septa, the inflammatory load, affected length ratio, maximum local evidence strength, and continuous distribution degree in the neighborhood of each structural instance are calculated, and full-slice-level features are obtained by using maximum value, mean, quantile, weighted summation, or attention-weighted methods.

5. The method as described in claim 1, characterized in that, The inflammatory activity assessment results include at least one of the following: no obvious interface inflammation-related inflammatory infiltration, presence of interface inflammation-related inflammatory infiltration, mild activity, moderate activity, severe activity, and continuous score, and include corresponding confidence information.

6. The method as described in claim 1, characterized in that, It also includes a quality control step; the quality control step generates a quality control result based on the effective tissue area, tissue sampling length, number of effective gate areas, image clarity, degree of staining abnormality, proportion of folded areas, proportion of blank areas or proportion of artifact areas; when the quality control result does not meet the preset conditions, a low confidence prompt or a manual review prompt is output.

7. A system for identifying and assessing inflammatory activity at the liver biopsy interface based on pathological structural constraints, characterized in that, include: Image acquisition module, used to acquire stained whole-section images of liver biopsy; The preprocessing module is used to extract the effective tissue region from the full slice image, divide the effective tissue region into multiple image blocks, and record the spatial coordinates of the effective tissue region and the image blocks in the coordinate system of the full slice image; The inflammation probability prediction module is used to predict the inflammation probability of the image block or the pixel region mapped from the image block using an inflammation recognition model, and generate an inflammation probability map. The pathological structure candidate map generation module is used to generate a pathological structure candidate map located in the same coordinate system as the inflammation probability map. The pathological structure candidate map includes a portal zone candidate region, a fibrous septum candidate region, and an interface zone candidate region. The interface zone candidate region is obtained by extending the boundary of the portal zone or the boundary of the fibrous septum towards the adjacent liver parenchyma. The spatial evidence generation module is used to generate a spatial evidence map of interface inflammation based on the spatial correlation between the inflammatory positive area and the candidate areas of the portal area, the candidate areas of the fibrous septum, or the candidate areas of the interface. Inflammation positive areas that are far from the candidate areas of the portal area, the candidate areas of the fibrous septum, and the candidate areas of the interface and do not satisfy the spatial correlation are identified as scattered inflammatory infiltration within the lobules, and the contribution of the inflammatory positive area to the spatial evidence map of interface inflammation is reduced. The evaluation output module is used to extract full-slice-level features from the spatial evidence map of the interface inflammation, and output the identification results of interface inflammation-related inflammatory infiltration and the evaluation results of inflammatory activity based on the full-slice-level features. as well as The visualization review module is used to display the inflammation probability heatmap, pathological structure candidate map, interface inflammation spatial evidence map, and inflammation activity assessment results; The interface inflammation spatial evidence map is obtained by fusing the inflammation probability map with spatial correlation weights. The spatial correlation weights include at least one of distance weights, structural neighborhood weights, and continuity weights. The distance weights reflect the distance from the location to the gate region or fiber septum boundary. The structural neighborhood weights reflect whether the location is located within the candidate region of the interface region or the neighborhood of the structural boundary. The continuity weights reflect the degree of continuous distribution of the inflammatory positive region along the structural boundary.

8. 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 method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 6.

Images