A multi-resolution pathological image progressive loading method

Abstract

The present application relates to pathological image processing technical field, disclose a kind of multi-resolution pathological image progressive loading method, including obtaining multiple original adjacent field image and scanning physical parameter, determine corresponding coarse transformation matrix;Based on coarse transformation matrix generates coarse splicing pathological image, and constructs the low-resolution pyramid base map of record field boundary, joint area and coarse transformation matrix index;Whether trigger on-demand fine splicing according to current display viewport position, zoom ratio and joint area position relationship judgment;After triggering, cut out the local data block associated with target joint area;End side based on coarse transformation matrix determines fine adjustment transformation parameter, and carries out deformation compensation and fusion rendering, generates fine splicing image.By constructing with field boundary, joint area and coarse transformation matrix index coarse splicing low-resolution pyramid base map, and when viewport meets trigger condition, execute local fine splicing, realize viewport cooperation, avoid waiting after whole image high-precision splicing can be browsed.

Description

Technical Field

[0001] This invention relates to the field of pathological image processing technology, specifically to a progressive loading method for multi-resolution pathological images. Background Technology

[0002] Pathological images refer to medical images formed after tissue sections are digitally acquired using microscopic scanning equipment. They typically contain microscopic structural information such as cell morphology, tissue boundaries, and staining distribution. They can be used for pathological slide reading, remote consultation, and digital pathology archiving. Progressive loading of multi-resolution pathological images refers to organizing large-size pathological images according to different resolution levels and gradually loading the corresponding image content based on the current viewport position and zoom level during user browsing. This allows users to first view the overall slice area and then gradually zoom in to observe local high-magnification areas. Current multi-resolution pathological image loading methods typically involve first performing global high-precision stitching and fusion of all original field-of-view images obtained from microscopic scanning on the server side, and then generating a multi-resolution pyramid based on the complete stitched image for client browsing.

[0003] However, with current technology, for ultra-large-scale pathological images or real-time scanning and reading scenarios under microscopes, the server needs to wait for the entire image to be fully stitched together globally with high precision before it can generate a viewable base map. This results in a long wait time for the first screen to load. At the same time, the area that doctors actually view at high magnification usually only occupies a small portion of the entire pathological image. Pre-stitching all the original field-of-view images with fine precision will require unnecessary calculations on a large number of unviewed areas, resulting in the ineffective use of the server's stitching computing power. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a progressive loading method for multi-resolution pathological images, solving the problem that pathological images must first undergo high-precision full-image stitching before they can be viewed, resulting in slow initial image loading and wasted server-side computing power.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a progressive loading method for multi-resolution pathological images, comprising: S1. Obtain multiple original adjacent field-of-view images formed by microscopic scanning and their scanning physical parameters, and determine the coarse transformation matrix of each original adjacent field-of-view image relative to the unified slice coordinate system based on the scanning physical parameters. S2. Perform coarse stitching on the multiple original adjacent field images according to the coarse transformation matrix to generate a coarse stitched pathological image, and construct a low-resolution pyramid base map based on the coarse stitched pathological image. The low-resolution pyramid base map records the field boundary, seam area and coarse transformation matrix index. S3. Obtain the position of the current display viewport and the current zoom level, and determine whether the on-demand fine splicing contact conditions are met based on the positional relationship between the current display viewport and the seam area and the current zoom level. S4. When the on-demand fine splicing contact condition is met, determine the target seam area covered by the current display viewport and its associated target original adjacent field of view image, and crop the corresponding local data block; S5. The end side receives the local data block, the corresponding coarse transformation matrix, and the target seam region position information. Based on the coarse transformation matrix, the local data block is initially aligned, and fine-tuning transformation parameters are determined in the overlapping areas on both sides of the target seam region. S6. Based on the fine-tuning transformation parameters, perform deformation compensation on the local data blocks within the current display viewport, and perform fusion rendering on the target seam area to generate a finely stitched display image corresponding to the current display viewport.

[0006] Preferably, determining the coarse transformation matrix of each original adjacent field-of-view image relative to the unified slice coordinate system based on the scanning physical parameters includes: Based on the mechanical coordinates and pixel calibration parameters corresponding to each original adjacent field of view image, the acquisition position of each original adjacent field of view image is converted to a unified slice coordinate system to obtain the initial translation parameters. Based on the scanning magnification, field size and nominal overlap width between adjacent original field images, the initial translation parameters are scaled and the overlap position is corrected to obtain the corrected translation parameters. Based on the corrected translation parameters, a coarse transformation matrix is ​​generated corresponding to each original adjacent field of view image, and an index relationship is established between the coarse transformation matrix and the corresponding original adjacent field of view image.

[0007] Preferably, the step of coarsely stitching the plurality of original adjacent field-of-view images according to the coarse transformation matrix to generate a coarsely stitched pathological image, and constructing a low-resolution pyramid base map based on the coarsely stitched pathological image, includes: Based on the coarse transformation matrix corresponding to each original adjacent field of view image, the multiple original adjacent field of view images are mapped to a unified slice coordinate system to form a coarsely stitched pathological image; Based on the overlapping area or adjacent boundary of adjacent original field-of-view images in the unified slice coordinate system, the seam region is determined, and the field-of-view boundary and coarse transformation matrix index corresponding to the seam region are recorded. The coarsely stitched pathological image is downsampled to generate a low-resolution pyramid base image, and the field of view boundary, the seam region, and the coarse transformation matrix index are preserved in the low-resolution pyramid base image.

[0008] Preferably, determining whether the on-demand fine-tuning trigger condition is met based on the positional relationship between the current display viewport and the seam area and the current zoom level includes: Map the position of the current display viewport to the unified slice coordinate system corresponding to the low-resolution pyramid base map to obtain the viewport mapping range; Determine whether the viewport mapping range covers the seam area or the distance between the viewport and the seam area is less than a preset distance threshold, and determine whether the current zoom level has reached a preset zoom level threshold; When the viewport mapping range covers the seam area or the distance between the viewport and the seam area is less than a preset distance threshold, and the current zoom level reaches a preset zoom level threshold, it is determined that the on-demand fine splicing trigger condition is met.

[0009] Preferably, when the on-demand fine-machining contact condition is met, determining the target seam area covered by the current display viewport and its associated original adjacent field-of-view image, and cropping the corresponding local data block, includes: The current display viewport that meets the on-demand fine splicing contact conditions is mapped to the unified slice coordinate system corresponding to the low-resolution pyramid base map, and the seam area within the viewport mapping range is determined as the target seam area. Based on the field of view boundary and coarse transformation matrix index corresponding to the target seam region, the original adjacent field of view images of the target associated on both sides of the target seam region are determined. Based on the mapping relationship between the current display viewport, the target seam region, and the target original adjacent field of view image, a local data block covering the overlapping area on both sides of the target seam region is cropped from the target original adjacent field of view image.

[0010] Preferably, the step of cropping a local data block from the original adjacent field-of-view image of the target, covering the overlapping areas on both sides of the target seam region, includes: Based on the reverse mapping position of the target seam region in the original adjacent field of view of the target, the basic cropping range in the original adjacent field of view of the target on both sides of the seam is determined respectively. Based on the boundary position of the current display viewport, the basic cropping range is extended by a preset width along the outside of the viewport to obtain an extended cropping range for local feature matching and fusion transition; According to the extended cropping range, local data blocks are cropped from the original adjacent field-of-view image of the target, and a transmission association is established between the local data blocks and the corresponding coarse transformation matrix and the target seam region location information.

[0011] Preferably, the terminal receives the local data block, the corresponding coarse transformation matrix, and the target seam region location information, performs initial alignment of the local data block based on the coarse transformation matrix, and determines fine-tuning transformation parameters within the overlapping regions on both sides of the target seam region, including: The edge receives the local data block, the corresponding coarse transformation matrix, and the target seam region position information, and maps the local data block to the edge rendering coordinate system based on the coarse transformation matrix to obtain the initially aligned local data block; Based on the target seam area location information, the overlapping area on both sides of the seam is determined in the initially aligned local data block, and gradient corner points, tissue edge points and texture block features are extracted from the overlapping area; Based on the gradient corner points, the tissue edge points, and the texture block features, matching point pairs are established between local data blocks on both sides of the seam, and fine-tuning transformation parameters are determined according to the matching point pairs.

[0012] Preferably, determining the fine-tuning transformation parameters based on the matching point pairs includes: The residual offset between local data blocks on both sides of the seam is calculated based on the matching point pairs, and matching point pairs whose offset direction or offset magnitude does not meet the consistency condition are removed. Based on the retained matching point pairs, the local translation and local rotation amounts are determined within a preset fine-tuning range; When the difference between the residual offsets at different locations reaches a preset deformation threshold, a mesh deformation is generated based on the residual offsets, and the local translation, the local rotation, and the mesh deformation are used as the fine-tuning transformation parameters.

[0013] Preferably, the step of performing deformation compensation on local data blocks within the current display viewport based on the fine-tuned transformation parameters, and performing fusion rendering on the target seam area to generate a finely stitched display image corresponding to the current display viewport includes: The display position of the local data block in the end-side rendering coordinate system is corrected according to the fine-tuning transformation parameters, so that the local data blocks on both sides of the target seam area can complete deformation compensation in the overlapping area. A fusion window is generated within the target seam area, and the fusion weights corresponding to the local data blocks on both sides of the seam are determined based on the distance from the pixels within the fusion window to the center line of the seam. Based on the fusion weight, the local data blocks after deformation compensation are weighted and fused for rendering to generate a finely stitched display image corresponding to the current display viewport.

[0014] Preferably, after generating the finely stitched display image corresponding to the current display viewport, the method further includes: When the current display viewport moves or scales, the changed display viewport position and the changed scaling factor are obtained, and the changed display viewport is mapped to the unified slice coordinate system corresponding to the low-resolution pyramid base map; Determine whether the changed display viewport covers the seam area where the fine-tuning transformation parameters have been determined. If it does, reuse the corresponding fine-tuning transformation parameters to perform deformation compensation and blending rendering on the local data blocks within the changed display viewport. If the changed display viewport covers the new seam area, then the new seam area is determined as the new target seam area, and the local data block clipping, end-side initial alignment, fine-tuning transformation parameter determination, and blending rendering are re-executed.

[0015] This invention provides a progressive loading method for multi-resolution pathological images. It has the following beneficial effects: 1. This invention constructs a coarsely stitched low-resolution pyramid base map with view boundaries, seam regions, and coarse transformation matrix indices, and performs local fine stitching only when the viewport meets the trigger conditions. This allows global coarse stitching and viewport-based fine stitching to be performed in tandem, avoiding the need to wait for the entire image to be stitched with high precision before browsing.

[0016] 2. After meeting the conditions for on-demand fine splicing, this invention only retrieves local data blocks of the original field of view on both sides of the seam area covered by the current display viewport, reducing the transmission and processing of irrelevant original image data and lowering the computational burden and resource consumption caused by full fine splicing on the server side.

[0017] 3. The present invention performs initial alignment of local data blocks on the end side based on a coarse transformation matrix, and determines fine-tuning transformation parameters by combining feature matching of overlapping areas on both sides of the seam. This can correct local seam misalignment in high magnification viewports and improve the display continuity of tissue boundaries and texture structures.

[0018] 4. This invention performs deformation compensation and fusion rendering within the target seam area, enabling a smooth transition between the images on both sides of the seam in the current viewport, reducing visible breaks caused by mechanical positioning errors, local offsets, or staining differences, and improving the quality of high-magnification observation of pathological images.

[0019] 5. This invention reuses the determined fine-tuning transformation parameters after the viewport moves or zooms, and re-triggers local fine stitching for the seam area newly entering the viewport, which not only ensures the stability of the seam display during continuous browsing, but also reduces repeated matching and calculation. Attached Figure Description

[0020] Figure 1 This is an architectural diagram of a progressive loading method for multi-resolution pathological images according to the present invention. Detailed Implementation

[0021] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see the appendix Figure 1 This invention provides a progressive loading method for multi-resolution pathological images, comprising: S1. Obtain multiple original adjacent field-of-view images formed by microscopic scanning and their scanning physical parameters, and determine the coarse transformation matrix of each original adjacent field-of-view image relative to the unified slice coordinate system based on the scanning physical parameters. Furthermore, based on the scanning physical parameters, the coarse transformation matrix of each original adjacent field-of-view image relative to the unified slice coordinate system is determined, including: Based on the mechanical coordinates and pixel calibration parameters corresponding to each original adjacent field of view image, the acquisition position of each original adjacent field of view image is converted to a unified slice coordinate system to obtain the initial translation parameters. Based on the scanning magnification, field size and nominal overlap width between adjacent original field images, the initial translation parameters are scaled and the overlap position is corrected to obtain the corrected translation parameters. Based on the corrected translation parameters, a coarse transformation matrix is ​​generated for each original adjacent field of view image, and an index relationship is established between the coarse transformation matrix and the corresponding original adjacent field of view image.

[0023] Specifically, after acquiring the original adjacent field-of-view images, the upper left reference point or scanning start point corresponding to the entire pathological slide can be used as the origin of the unified slide coordinate system. Based on the mechanical coordinates of each original adjacent field-of-view image recorded by the scanner, the physical displacement of each field-of-view image relative to the origin can be determined. Since mechanical coordinates are usually expressed in units of length, while subsequent coarse stitching needs to be completed in image pixel coordinates, the mechanical displacement can be converted into the corresponding pixel displacement by combining the pixel calibration parameters of the scanner. This yields the initial translation parameters of each original adjacent field-of-view image in the unified slide coordinate system. Through this process, the original field-of-view images that were originally scattered at various acquisition positions can be uniformly mapped to the same coordinate reference, providing an initial position basis for subsequent rapid coarse stitching. After obtaining the initial translation parameters, the initial position of each field image can be further corrected by combining the scanning magnification, field size, and the pre-set nominal overlap width between adjacent original field images. Specifically, the number of pixels corresponding to a unit physical length may be different under different scanning magnifications. Therefore, it is necessary to standardize the initial translation parameters according to the scanning magnification. At the same time, considering that adjacent field images usually retain a certain overlap area during acquisition, the theoretical distance between adjacent field images can be corrected according to the field size and nominal overlap width to avoid cumulative offset when relying solely on mechanical coordinates. For example, when the field width of two adjacent field images in the scanning direction is 2048 pixels and the nominal overlap width is 128 pixels, the theoretical adjacent displacement of the two in this direction can be corrected by 1920 pixels, thereby obtaining a corrected translation parameter that is closer to the subsequent coarse stitching requirements. Through the above processing, the consistency between the coarse positioning result and the actual field arrangement can be improved without performing high-overhead image feature matching. After obtaining the corrected translation parameters, they can be written into the coarse transformation matrix corresponding to each original adjacent field of view image. This matrix describes the initial geometric transformation relationship required when each field of view image is mapped from its own coordinate system to the unified slice coordinate system. For scenarios used only for coarse stitching positioning, the coarse transformation matrix can at least include the corrected horizontal and vertical translation amounts. When it is necessary to retain the magnification conversion relationship, it can also include the scale parameter. Subsequently, an index relationship is established between each coarse transformation matrix and its corresponding original adjacent field of view image identifier. This allows the server to reverse-search the relevant original field of view image and its initial mapping relationship when locating the seam area based on the viewport position. By establishing this index relationship, the low-resolution coarse stitching base map can not only be used for browsing and display, but also serve as the spatial index basis for quickly retrieving the original local data during the subsequent on-demand fine stitching process.

[0024] S2. Perform coarse stitching on multiple original adjacent field images according to the coarse transformation matrix to generate a coarse stitched pathological image. Construct a low-resolution pyramid base map based on the coarse stitched pathological image. The low-resolution pyramid base map records the field boundary, seam area and coarse transformation matrix index. Furthermore, multiple original adjacent field-of-view images are coarsely stitched together based on the coarse transformation matrix to generate a coarsely stitched pathological image. A low-resolution pyramid base map is then constructed based on this coarsely stitched pathological image, including: Based on the coarse transformation matrix corresponding to each original adjacent field of view image, multiple original adjacent field of view images are mapped to a unified slice coordinate system to form a coarsely stitched pathological image. Based on the overlapping area or adjacent boundary of adjacent original field-of-view images in the unified slice coordinate system, determine the seam region and record the field-of-view boundary and coarse transformation matrix index corresponding to the seam region. The coarsely stitched pathological images are downsampled to generate a low-resolution pyramid base map, and the field of view boundary, seam region and coarse transformation matrix index are preserved in the low-resolution pyramid base map.

[0025] Specifically, after obtaining the coarse transformation matrix corresponding to each original adjacent field of view image, each original adjacent field of view image can be mapped from its own image coordinate system to a unified slice coordinate system according to the coarse transformation matrix. The server allocates the corresponding placement area for each original adjacent field of view image in the unified slice coordinate system and completes the rapid arrangement according to the corrected translation parameters or scale parameters, thereby forming a coarse stitched pathological image. This coarse stitched pathological image allows for slight misalignment at local seams. Its purpose is not to be directly used as the final high-magnification diagnostic image, but to be used to quickly form a viewable overall slice base map. For example, when multiple field of view images are acquired in row and column order during microscopic scanning, they can be quickly placed in the corresponding row and column positions in the unified slice coordinate system according to the coarse transformation matrix corresponding to each field of view image, thereby obtaining the coarse stitching result without performing full feature matching and global fusion. During the process of forming a coarsely stitched pathological image, the seam area can be further determined based on the spatial relationship between adjacent original field images in a unified slice coordinate system. For adjacent field images that overlap, the overlapping area can be used as the seam area. For adjacent field images that only form an adjacency relationship, the adjacent boundary between the two and the preset width range on both sides can be used as the seam area. After determining the seam area, the server records the field boundaries on both sides corresponding to the seam area and associates them with the coarse transformation matrix index of the corresponding original adjacent field images. Through this processing, when the user's viewport covers a certain seam area, the original field images and corresponding coarse transformation matrix associated on both sides can be quickly located according to the seam area index, avoiding the need to re-traverse the entire coarsely stitched pathological image. After completing the coarse stitching of the pathological image and recording of the seam area, the coarse stitched pathological image can be downsampled step by step to generate a low-resolution pyramid base map. The low-resolution pyramid base map is used for the client's first screen loading and zoom browsing, allowing users to obtain the overall view of the pathological slide before the global high-precision stitching is completed. During the downsampling process, the field boundary, seam area, and coarse transformation matrix index are not discarded as ordinary pixel information, but are retained in the low-resolution pyramid base map in the form of coordinate mapping or metadata index. Thus, the low-resolution pyramid base map not only serves as a browsing and positioning tool, but also serves as an index base map for subsequent viewport-driven on-demand fine stitching, enabling the system to quickly retrieve the corresponding original local data and perform end-side fine-tuning fusion when the user enters the high-magnification seam viewport.

[0026] S3. Obtain the position and zoom level of the current display viewport, and determine whether the conditions for on-demand fine splicing are met based on the positional relationship between the current display viewport and the seam area and the current zoom level. Furthermore, based on the current positional relationship between the display viewport and the seam area, and the current zoom level, it is determined whether the conditions for on-demand fine-tuning are met, including: Map the current viewport position to the unified slice coordinate system corresponding to the low-resolution pyramid base map to obtain the viewport mapping range; Determine whether the viewport mapping range covers the seam area or the distance between the viewport and the seam area is less than a preset distance threshold, and determine whether the current zoom level has reached a preset zoom level threshold. When the viewport mapping range covers the seam area or the distance between the viewport and the seam area is less than a preset distance threshold, and the current zoom level reaches a preset zoom level threshold, the on-demand fine splicing trigger condition is determined to be met.

[0027] Specifically, when browsing a low-resolution pyramid base map, the client can obtain the current viewport's position range in the client's display coordinate system and the current zoom level in real time. Because there is a zoom level difference between the client's display coordinate system and the unified chip coordinate system used by the server to store the seam area, the current viewport's position can be converted back to the unified chip coordinate system based on the zoom ratio corresponding to the current pyramid level, thus obtaining the viewport mapping range. This process ensures that the user's actual observation area and the seam area recorded by the server are judged under the same coordinate reference, avoiding false triggering or missed triggering of local fine-tuning due to inconsistencies in coordinates between different resolution levels.

[0028] After obtaining the viewport mapping range, the system can determine whether the viewport mapping range has covered any seam area, or whether the minimum distance between the viewport mapping range and the seam area is less than a preset distance threshold; simultaneously, it can determine whether the current zoom level has reached a preset zoom level threshold. Specifically, the following trigger determination formula can be used: ; in, For the first The current display viewport shows the on-demand precision splicing contact determination value. This is an indicator function that takes the value 1 when the condition within the parentheses is true, and 0 otherwise. For the first The zoom level corresponding to the current display viewport at any given time. To trigger a preset magnification threshold for local fine stitching, For the first The viewport mapping range of the current display viewport in the unified chip coordinate system at any given time. For the first One seam area; This is the minimum spatial distance between the viewport mapping area and the seam area. When there is overlap between the two, this minimum spatial distance is 0. The above determination method is a preset distance threshold used to determine whether the current display viewport is adjacent to the seam area. The system only starts local fine splicing processing when the user enters a high-magnification observation state that may be affected by the seam, and does not consume end-side matching and fusion resources when browsing the global view at low magnification or far away from the seam area. When the judgment value is triggered When the value is 1, it can be determined that the current display viewport meets the conditions for triggering on-demand fine stitching, and proceed to subsequent target seam area determination, original adjacent local field local data cropping, and end-side fine-tuning fusion processing. For example, if the user views the entire slice at a low magnification, even if the current viewport range covers multiple seam areas, fine stitching will not be triggered because the zoom level has not reached the preset magnification threshold. If the user zooms in to a high magnification but the current viewport is far from the seam area, fine stitching will not be triggered either. Only when the user zooms in to a high magnification and the current viewport covers or is close to a certain seam area will the system treat that seam area as the object of subsequent processing. Thus, this implementation can limit the computationally intensive fine stitching processing to the local area that the user actually needs to observe, while taking into account both the rapid output of the coarse stitching base map and the display continuity of the high magnification seam area.

[0029] S4. When the on-demand fine splicing contact conditions are met, determine the target seam area covered by the current display viewport and its associated target original adjacent field of view image, and crop the corresponding local data block; Furthermore, when the on-demand fine-machining contact condition is met, the target seam area covered by the current display viewport and its associated original adjacent field-of-view images are determined, and the corresponding local data blocks are cropped, including: Map the current display viewport that meets the conditions for on-demand fine splicing to the unified slice coordinate system corresponding to the low-resolution pyramid base map, and determine the seam area within the viewport mapping range as the target seam area. Based on the field of view boundary and coarse transformation matrix index corresponding to the target seam region, determine the original adjacent field of view images of the target on both sides of the target seam region. Based on the mapping relationship between the current display viewport, the target seam area, and the original adjacent field of view image of the target, a local data block covering the overlapping area on both sides of the target seam area is cropped from the original adjacent field of view image of the target.

[0030] Specifically, after meeting the conditions for on-demand fine stitching, the server can first read the position range of the current display viewport under the current zoom level of the client, and convert the current display viewport into the viewport mapping range in the unified chip coordinate system according to the hierarchical ratio between the low-resolution pyramid base map and the unified chip coordinate system. Then, the viewport mapping range is spatially intersected with each seam area pre-recorded in the low-resolution pyramid base map. The seam area located within the viewport mapping range is determined as the target seam area. If the viewport mapping range covers multiple seam areas at the same time, a list of target seam areas can be established separately, and the processing order is determined according to the distance between the seam area and the center of the viewport or the visible area of ​​the seam area in the viewport. In this way, the system can limit the subsequent fine stitching objects to the seam positions actually observed by the current user, rather than uniformly performing fine stitching processing on all seams in the entire pathological image. After identifying the target seam region, the field of view boundary and coarse transformation matrix index recorded in the low-resolution pyramid base image can be further accessed. Since each seam region is formed by the overlapping area or adjacent boundary of adjacent original field of view images, the original field of view identifiers on both sides of the seam region can be determined based on the field of view boundary. The coarse transformation relationship of each original field of view image relative to the unified chip coordinate system can be read based on the coarse transformation matrix index, thereby determining the target original adjacent field of view images associated with both sides of the target seam region. For example, when the target seam region is located between the i-th field of view image and the (i+1)-th field of view image, the server can directly locate the two original field of view images and their coarse transformation matrix based on the field of view boundary index corresponding to the seam region, avoiding a re-search in all original field of view images. After determining the target's original adjacent field of view image, local data blocks can be cropped from the corresponding original field of view image based on the mapping relationship between the current display viewport, the target seam area, and the target's original adjacent field of view image. Specifically, the area to be processed, jointly formed by the current display viewport and the target seam area, can be reverse-mapped from the unified chip coordinate system to the image coordinate system of the target's original adjacent field of view image. The reverse-mapped area is then used as the basic cropping range to obtain local data blocks covering the overlapping areas on both sides of the target seam area. This cropping range can be represented as: ; in, This represents the range of a local data block cropped from the original neighboring field-of-view image of the i-th target. This represents the inverse transform of the coarse transform matrix corresponding to the original adjacent field-of-view image of the i-th target. This indicates the viewport mapping range of the current display viewport in the unified chip coordinate system. This indicates the j-th target seam area currently displayed in the viewport. This indicates the area to be clipped, which is jointly defined by the current viewport and the target seam area. Through clipping, the data down to the edge is no longer ordinary display tiles, but original local data directly related to the target seam within the current viewport. This provides an image basis that is closer to the original acquisition state for subsequent edge feature matching, fine-tuning compensation, and fusion rendering.

[0031] Furthermore, local data blocks covering the overlapping areas on both sides of the target seam region are cropped from the original adjacent field-of-view image of the target, including: Based on the reverse mapping position of the target seam region in the original adjacent field of view image of the target, the basic cropping range in the original adjacent field of view image of the target on both sides of the seam is determined respectively. Based on the current viewport boundary position, the basic clipping range is extended by a preset width along the outside of the viewport to obtain an extended clipping range for local feature matching and fusion transition; Local data blocks are extracted from the original adjacent field-of-view image of the target according to the extended cropping range, and the local data blocks are associated with the corresponding coarse transformation matrix and the location information of the target seam area.

[0032] Specifically, based on the position of the target seam region in the unified slice coordinate system, the coarse transformation matrix corresponding to the original adjacent field-of-view images of the target on both sides of the seam can be used for inverse mapping to convert the target seam region into the original image coordinate system of each target's original adjacent field-of-view image. Since the same seam region is usually formed by the overlapping area or adjacent boundary of two adjacent field-of-view images, after inverse mapping, the local position corresponding to the target seam region in the images on both sides of the seam can be obtained respectively, and the basic clipping range is determined based on the local position. The basic clipping range at least covers the overlapping image content used for registration on both sides of the target seam region, so that the end side can obtain sufficient image information on both sides of the seam after receiving the local data block, without having to request the complete original field-of-view image again. After determining the basic cropping range, the basic cropping range can be extended by a preset width along the outer edge of the viewport, based on the current viewport boundary position, to obtain the extended cropping range. The preset width is used to retain redundant image content outside the seam neighborhood, so that the end side has a sufficient feature search range when performing local feature matching, and forms a smooth transition region during fusion rendering. This extended cropping range can be expressed as: ; in, Indicates from the first The range of local data blocks cropped from the original adjacent field-of-view images of each target. Indicates the first The inverse transform of the coarse transform matrix corresponding to the original adjacent field-of-view images of each target. This indicates the viewport mapping range of the currently displayed viewport in the uniform slice coordinate system. Indicates the target seam area. This indicates an expansion of a preset width along the region boundary. Extended operations, This indicates the preset outward expansion width used for local feature matching and fusion transition. For example, when the current display viewport just covers the seam boundary of two adjacent field images, the server can reserve a certain width of outward expansion area on both sides of the seam so that the end side will not lack matching tissue edges or texture blocks due to the cropping boundary being too narrow when calculating the residual offset in the later stage. After obtaining the extended cropping range, the server extracts local data blocks from the original adjacent field-of-view images of the target on both sides of the seam according to the range. The extracted local data blocks are then linked to the corresponding coarse transformation matrix, the target seam region location information, and the field-of-view identifier for transmission. During transmission, the local data block can be sent to the endpoint along with its original field-of-view image identifier, the corresponding coarse transformation matrix index, the target seam region number, and the viewport coordinates. This allows the endpoint to directly complete initial alignment, feature matching, and fusion rendering based on the transmission linkage. As a result, the data sent is limited to the local range related to the current viewport and the target seam, which reduces the amount of original image data transmitted while ensuring that the endpoint obtains the original images of both sides of the seam required for local fine stitching.

[0033] S5. The end side receives local data blocks, the corresponding coarse transformation matrix, and the target seam region position information. Based on the coarse transformation matrix, the local data blocks are initially aligned, and the fine-tuning transformation parameters are determined in the overlapping areas on both sides of the target seam region. Furthermore, the endpoint receives local data blocks, the corresponding coarse transformation matrix, and the target seam region location information. Based on the coarse transformation matrix, it performs initial alignment of the local data blocks and determines fine-tuning transformation parameters within the overlapping regions on both sides of the target seam region, including: The end side receives local data blocks, the corresponding coarse transformation matrix, and the target seam region position information, and maps the local data blocks to the end side rendering coordinate system based on the coarse transformation matrix to obtain the initially aligned local data blocks; Based on the location information of the target seam area, the overlapping areas on both sides of the seam are determined in the initially aligned local data blocks, and gradient corner points, tissue edge points and texture block features are extracted from the overlapping areas. Based on gradient corner points, tissue edge points, and texture block features, matching point pairs are established between local data blocks on both sides of the seam, and fine-tuning transformation parameters are determined according to the matching point pairs.

[0034] Specifically, after receiving the local data blocks, the corresponding coarse transformation matrix, and the target seam region location information from the server, the edge device can first project each local data block onto the edge device's rendering coordinate system according to the coarse transformation matrix. This allows the local data blocks on both sides of the seam to be displayed and aligned according to the initial positional relationship during the server's coarse stitching. This initial alignment does not require the complete elimination of seam errors, but rather provides a smaller search range for subsequent local feature matching, avoiding the edge device from re-performing global registration within a larger image range. For example, when the target seam region is located on both sides of an adjacent first field of view image and a second field of view image, the edge device can first place the local data blocks on both sides into the same rendering coordinate system according to their corresponding coarse transformation matrices, so that their overlapping areas roughly correspond. After initial alignment, the end side can extract the overlapping area on both sides of the seam from the local data block after initial alignment based on the location information of the target seam area, and extract image features for local matching within the overlapping area. Specifically, gradient corner points with significant gray-scale or brightness changes can be extracted from the overlapping area to characterize the location of local texture intersections, and tissue edge points can be extracted to characterize the continuous location of structures such as gland edges, tissue contours, or cell cluster boundaries. At the same time, texture block features with stable texture distribution are selected to supplement the matching basis in areas with staining changes or weak edges. By using point features, edge features, and local texture blocks simultaneously, the matching stability of both sides of the seam can be improved in cases of complex pathological tissue structures, uneven staining, or repetitive local textures. After obtaining gradient corner points, tissue edge points, and texture block features, the end-view can establish matching point pairs between local data blocks on both sides of the seam. Specifically, candidate matching relationships can be filtered based on the similarity of feature descriptors, and then candidate point pairs that are significantly deviated can be eliminated by combining the relative position constraints of the matching points on both sides of the seam. This yields a set of matching point pairs for fine-tuning transformation calculations. For texture block features, the center point of the texture block can be used as the matching position in subsequent calculations. That is, the end-view can obtain the residual positional relationship between local data blocks on both sides of the seam based on the initial alignment provided by the coarse transformation matrix. This provides a data basis for subsequently determining local translation, local rotation, or mesh deformation, thereby enabling on-demand fine-tuning only for the target seam area within the current display viewport, without having to perform global fine stitching on the entire pathological image.

[0035] Furthermore, the fine-tuning transformation parameters are determined based on the matching point pairs, including: Calculate the residual offset between local data blocks on both sides of the seam based on the matching point pairs, and remove matching point pairs whose offset direction or offset magnitude does not meet the consistency condition; Based on the retained matching point pairs, the local translation and local rotation amounts are determined within a preset fine-tuning range; When the difference between the residual offsets at different locations reaches the preset deformation threshold, the mesh deformation is generated based on the residual offsets, and the local translation, local rotation, and mesh deformation are used as fine-tuning transformation parameters.

[0036] Specifically, after establishing matching point pairs between local data blocks on both sides of the seam, the initial mapping relationship formed by the coarse transformation matrix can be used as a benchmark to calculate the residual offset corresponding to each matching point pair. Abnormal matching point pairs are then eliminated based on the consistency of the residual offsets. Specifically, the first matching point pair in the local data block on one side of the seam can be set as follows: The feature points are The feature points that match it on the other side of the seam are The initial mapping relationship formed by the coarse transformation matrix is: Then the first The residual offset of a group of matched point pairs can be expressed as: ; The mean of the residual offset can be expressed as: ; No. The offset consistency error of a group of matched point pairs can be expressed as: ; when When, retain the matching point pair in that group. When that time, remove the matching point pair in the group, where, For the first The residual offset of the group-matched point pairs based on the coarse transform. The mean of the residual offset. This represents the initial number of matched point pairs. This represents the degree of deviation of the set of matched point pairs relative to the overall residual offset trend. By setting a consistency threshold, this process can remove erroneous matching points caused by uneven staining, repetitive tissue texture, or local blurring, so that subsequent fine-tuning of transformation parameters is based on a more stable matching relationship. After retaining the matching point pairs that meet the consistency condition, the local translation and local rotation between local data blocks on both sides of the seam can be solved within a preset fine-tuning range. Since the coarse transformation matrix has already provided the initial alignment relationship, the end side does not need to perform large-scale global registration again, but only needs to correct the residual offset within a small parameter range. Specifically, the fine-tuning transformation parameters can be solved as follows: ; in, The fine-tuning transformation parameters are for the target seam area. For the local fine-tuning parameters to be solved, To preset the fine-tuning range, This refers to the number of valid matching pairs retained after removing outlier matches. This refers to the fine-tuning transformation determined by local fine-tuning parameters. The preset fine-tuning range can be limited to: ; in, and These represent the horizontal and vertical local translation amounts, respectively. This is the local rotation amount. , and These are the corresponding preset fine-tuning limits. For example, when the two sides of a seam in the coarse stitching base image are roughly aligned, but there are still several pixels of horizontal misalignment when displayed at high magnification, the end side can solve the local translation amount within the preset translation range, and further determine the local rotation amount based on a small amount of rotation error, thereby making a light correction to the seam in the current viewport. When there are significant differences in the residual offsets at different locations, it indicates that the seam area may not only have overall translational or rotational deviations, but also non-uniform offsets caused by minor vibrations of the scanning platform, local tissue deformation, or lens distortion. In this case, the degree of dispersion between the residual offsets can be used to determine whether to enable mesh deformation compensation. Specifically, the dispersion of the residual offsets within the target seam area can be calculated: ; when At that time, it is determined that there is a local non-uniform residual offset in the target seam area, and a mesh deformation is generated based on the residual offset of each effective matching point pair. For the position to be compensated within the target seam area... Its grid deformation can be expressed as: ; in: ; in, For the first The dispersion of residual offset within each target seam area To preset the deformation threshold, For position The grid shape at that location, For the first A valid matching point pair at position Influence weight, To avoid positive constants with a denominator of zero, the end side can use only local translation and local rotation for lightweight correction when the residual offset is basically consistent. When the residual offset shows local inconsistency, mesh deformation is introduced. This ensures the continuity of the current viewport seam while avoiding unnecessary high-overhead non-rigid registration. Finally, the obtained local translation, local rotation, and mesh deformation generated under the triggering conditions are used as fine-tuning transformation parameters for subsequent deformation compensation and blending rendering.

[0037] S6. Based on the fine-tuning transformation parameters, perform deformation compensation on the local data blocks in the current display viewport, and perform fusion rendering on the target seam area to generate a finely stitched display image corresponding to the current display viewport.

[0038] Furthermore, based on fine-tuning transformation parameters, deformation compensation is performed on local data blocks within the current display viewport, and the target seam area is blended and rendered to generate a finely stitched display image corresponding to the current display viewport, including: The display position of the local data block in the end-side rendering coordinate system is corrected according to the fine-tuning transformation parameters, so that the local data blocks on both sides of the target seam area can complete deformation compensation in the overlapping area. A fusion window is generated within the target seam area, and the fusion weights corresponding to the local data blocks on both sides of the seam are determined based on the distance from the pixels within the fusion window to the center line of the seam. Based on the fusion weight, the local data blocks after deformation compensation are weighted and fused for rendering to generate a finely stitched display image corresponding to the current display viewport.

[0039] Specifically, after the fine-tuning transformation parameters are determined, the end-side can apply these parameters to local data blocks within the current display viewport to correct the display position of the local data blocks in the end-side rendering coordinate system. For local data blocks on one side of the target seam area, their display coordinates can be updated based on the local translation and local rotation. When the fine-tuning transformation parameters include grid deformation, the pixel coordinates at different positions in the seam neighborhood can be further compensated according to the deformation corresponding to each grid node. This processing only applies to the local data blocks covered by the current display viewport and does not change the global coordinate relationship of the coarsely stitched pathological image, thereby correcting the local misalignment of the current seam area during the real-time rendering process on the end-side. After deformation compensation is completed, a fusion window can be generated within the target seam area. The fusion window can be formed by expanding a preset width to both sides along the seam center line. Its width can be determined according to the overlap width of the seam area, the current scaling factor, or the outward expansion width of the local data block. The end side calculates the distance from each pixel to the seam center line within the fusion window, and determines the fusion weight corresponding to the local data blocks on both sides of the seam based on the distance. This allows pixels closer to the main area of ​​the local data block on one side to adopt more of the pixel values ​​of that local data block, while pixels closer to the seam center line smoothly transition between the local data blocks on both sides according to the distance change. By setting the fusion window, the visible breaks in the seam area caused by brightness differences, color differences, or local texture differences can be reduced. The edge device performs weighted fusion rendering on local data blocks after deformation compensation based on fusion weights to generate a finely stitched display image corresponding to the current display viewport. Specifically, in non-seam areas, the edge device can directly use the pixel content of the corresponding local data block for display. In the fusion window, the rendered pixel value is calculated according to the fusion weight of the local data blocks on both sides of the seam, and the fused pixel result is output to the current display viewport. For example, when a user views the tissue edge between two adjacent field images at high magnification, the edge device first eliminates the local misalignment of the tissue edge on both sides of the seam through fine-tuning transformation, and then performs smooth fusion in the seam neighborhood to make the tissue structure in the current viewport continuously displayed. Thus, without waiting for the server to complete the global fine stitching of the entire pathological image, the edge device can obtain the local fine stitching display result in the high magnification area actually observed by the user.

[0040] Furthermore, after generating the finely stitched display image corresponding to the current display viewport, it also includes: When the current display viewport moves or scales, obtain the changed display viewport position and the changed scaling factor, and map the changed display viewport to the unified tile coordinate system corresponding to the low-resolution pyramid base map; Determine whether the changed display viewport covers the seam area where the fine-tuning transformation parameters have been determined. If it does, reuse the corresponding fine-tuning transformation parameters to perform deformation compensation and blending rendering on the local data blocks within the changed display viewport. If the changed display viewport covers the new seam area, the new seam area will be identified as the new target seam area, and the local data block clipping, edge initial alignment, fine-tuning transformation parameter determination, and blending rendering will be re-executed.

[0041] Specifically, after generating the finely stitched display image corresponding to the current display viewport, the client can continuously monitor the translation, scaling, or layer switching operations of the display viewport. When a movement or scaling change of the current display viewport is detected, the changed position of the display viewport and the changed scaling factor are obtained. Based on the changed pyramid layer ratio, the changed display viewport is remapped to the unified slice coordinate system corresponding to the low-resolution pyramid base map. Through this remapping, the client or server can determine the spatial relationship between the changed display viewport and the processed seam area and the newly entered seam area under a unified coordinate reference, thereby ensuring that the seam processing results during the continuous movement of the viewport can be accurately tracked. After obtaining the changed display viewport mapping range, the spatial coverage of the mapping range and the seam area with the already determined fine-tuned transformation parameters can be judged. If the changed display viewport still covers the seam area that has been fine-tuned, or is still within the preset neighborhood of the seam area, the end side can reuse the fine-tuned transformation parameters corresponding to the seam area to continue to perform deformation compensation and blending rendering on the relevant local data blocks in the changed display viewport. For example, when the user slowly drags the viewport along the same seam area, there is usually an overlap between adjacent display viewports. The end side can use the local translation amount, local rotation amount, or mesh deformation that has been obtained from the previous viewport, and only perform supplementary rendering on the newly added local data entering the edge of the viewport, thereby reducing repeated feature matching and repeated parameter solving, and improving the display response speed during continuous browsing. If the changed display viewport covers a new seam area, and the corresponding fine-tuning transformation parameters for this new seam area have not yet been established, then this new seam area can be identified as the new target seam area. The associated target original adjacent field of view image is then re-determined according to the aforementioned process. Local data blocks covering the overlapping areas on both sides of the seam area are cropped. Subsequently, initial alignment, feature matching, fine-tuning transformation parameter determination, and fusion rendering are performed on the end side. Through this update method, new on-demand fine stitching processing can be automatically triggered when the viewport moves to a new seam position, and existing results can be reused when the viewport is still near the processed seam. This ensures the continuity of high-magnification seam display while reducing the amount of redundant calculations on the end side and redundant data retrieval on the server side.

[0042] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A progressive loading method for multi-resolution pathological images, characterized in that, include: S1. Obtain multiple original adjacent field-of-view images formed by microscopic scanning and their scanning physical parameters, and determine the coarse transformation matrix of each original adjacent field-of-view image relative to the unified slice coordinate system based on the scanning physical parameters. S2. Perform coarse stitching on the multiple original adjacent field images according to the coarse transformation matrix to generate a coarse stitched pathological image, and construct a low-resolution pyramid base map based on the coarse stitched pathological image. The low-resolution pyramid base map records the field boundary, seam area and coarse transformation matrix index. S3. Obtain the position of the current display viewport and the current zoom level, and determine whether the on-demand fine splicing contact conditions are met based on the positional relationship between the current display viewport and the seam area and the current zoom level. S4. When the on-demand fine splicing contact condition is met, determine the target seam area covered by the current display viewport and its associated target original adjacent field of view image, and crop the corresponding local data block; S5. The end side receives the local data block, the corresponding coarse transformation matrix, and the target seam region position information. Based on the coarse transformation matrix, the local data block is initially aligned, and fine-tuning transformation parameters are determined in the overlapping areas on both sides of the target seam region. S6. Based on the fine-tuning transformation parameters, perform deformation compensation on the local data blocks within the current display viewport, and perform fusion rendering on the target seam area to generate a finely stitched display image corresponding to the current display viewport.

2. The progressive loading method for multi-resolution pathological images according to claim 1, characterized in that, The determination of the coarse transformation matrix of each original adjacent field-of-view image relative to the unified slice coordinate system based on the scanning physical parameters includes: Based on the mechanical coordinates and pixel calibration parameters corresponding to each original adjacent field of view image, the acquisition position of each original adjacent field of view image is converted to a unified slice coordinate system to obtain the initial translation parameters. Based on the scanning magnification, field size and nominal overlap width between adjacent original field images, the initial translation parameters are scaled and the overlap position is corrected to obtain the corrected translation parameters. Based on the corrected translation parameters, a coarse transformation matrix is ​​generated corresponding to each original adjacent field of view image, and an index relationship is established between the coarse transformation matrix and the corresponding original adjacent field of view image.

3. The progressive loading method for multi-resolution pathological images according to claim 1, characterized in that, The step of coarsely stitching the multiple original adjacent field-of-view images according to the coarse transformation matrix to generate a coarsely stitched pathological image, and constructing a low-resolution pyramid base map based on the coarsely stitched pathological image, includes: Based on the coarse transformation matrix corresponding to each original adjacent field of view image, the multiple original adjacent field of view images are mapped to a unified slice coordinate system to form a coarsely stitched pathological image; Based on the overlapping area or adjacent boundary of adjacent original field-of-view images in the unified slice coordinate system, the seam region is determined, and the field-of-view boundary and coarse transformation matrix index corresponding to the seam region are recorded. The coarsely stitched pathological image is downsampled to generate a low-resolution pyramid base image, and the field of view boundary, the seam region, and the coarse transformation matrix index are preserved in the low-resolution pyramid base image.

4. The progressive loading method for multi-resolution pathological images according to claim 1, characterized in that, The step of determining whether the on-demand fine-tuning contact triggering condition is met based on the positional relationship between the current display viewport and the seam area and the current zoom level includes: Map the position of the current display viewport to the unified slice coordinate system corresponding to the low-resolution pyramid base map to obtain the viewport mapping range; Determine whether the viewport mapping range covers the seam area or the distance between the viewport and the seam area is less than a preset distance threshold, and determine whether the current zoom level has reached a preset zoom level threshold; When the viewport mapping range covers the seam area or the distance between the viewport and the seam area is less than a preset distance threshold, and the current zoom level reaches a preset zoom level threshold, it is determined that the on-demand fine splicing trigger condition is met.

5. The progressive loading method for multi-resolution pathological images according to claim 1, characterized in that, When the on-demand precision stitching contact condition is met, the target seam area covered by the current display viewport and its associated original adjacent field-of-view image are determined, and the corresponding local data block is cropped, including: The current display viewport that meets the on-demand fine splicing contact conditions is mapped to the unified slice coordinate system corresponding to the low-resolution pyramid base map, and the seam area within the viewport mapping range is determined as the target seam area. Based on the field of view boundary and coarse transformation matrix index corresponding to the target seam region, the original adjacent field of view images of the target associated on both sides of the target seam region are determined. Based on the mapping relationship between the current display viewport, the target seam region, and the target original adjacent field of view image, a local data block covering the overlapping area on both sides of the target seam region is cropped from the target original adjacent field of view image.

6. The progressive loading method for multi-resolution pathological images according to claim 5, characterized in that, The step of cropping a local data block from the original adjacent field-of-view image of the target, covering the overlapping areas on both sides of the target seam region, includes: Based on the reverse mapping position of the target seam region in the original adjacent field of view of the target, the basic cropping range in the original adjacent field of view of the target on both sides of the seam is determined respectively. Based on the boundary position of the current display viewport, the basic cropping range is extended by a preset width along the outside of the viewport to obtain an extended cropping range for local feature matching and fusion transition; According to the extended cropping range, local data blocks are cropped from the original adjacent field-of-view image of the target, and a transmission association is established between the local data blocks and the corresponding coarse transformation matrix and the target seam region location information.

7. The progressive loading method for multi-resolution pathological images according to claim 1, characterized in that, The endpoint receives the local data block, the corresponding coarse transformation matrix, and the target seam region position information. Based on the coarse transformation matrix, it performs initial alignment of the local data block and determines fine-tuning transformation parameters within the overlapping regions on both sides of the target seam region, including: The edge receives the local data block, the corresponding coarse transformation matrix, and the target seam region position information, and maps the local data block to the edge rendering coordinate system based on the coarse transformation matrix to obtain the initially aligned local data block; Based on the target seam area location information, the overlapping area on both sides of the seam is determined in the initially aligned local data block, and gradient corner points, tissue edge points and texture block features are extracted from the overlapping area; Based on the gradient corner points, the tissue edge points, and the texture block features, matching point pairs are established between local data blocks on both sides of the seam, and fine-tuning transformation parameters are determined according to the matching point pairs.

8. The progressive loading method for multi-resolution pathological images according to claim 7, characterized in that, The step of determining the fine-tuning transformation parameters based on the matching point pairs includes: The residual offset between local data blocks on both sides of the seam is calculated based on the matching point pairs, and matching point pairs whose offset direction or offset magnitude does not meet the consistency condition are removed. Based on the retained matching point pairs, the local translation and local rotation amounts are determined within a preset fine-tuning range; When the difference between the residual offsets at different locations reaches a preset deformation threshold, a mesh deformation is generated based on the residual offsets, and the local translation, the local rotation, and the mesh deformation are used as the fine-tuning transformation parameters.

9. The progressive loading method for multi-resolution pathological images according to claim 1, characterized in that, The step of performing deformation compensation on local data blocks within the current display viewport based on the fine-tuned transformation parameters, and performing blending rendering on the target seam area to generate a finely stitched display image corresponding to the current display viewport includes: The display position of the local data block in the end-side rendering coordinate system is corrected according to the fine-tuning transformation parameters, so that the local data blocks on both sides of the target seam area can complete deformation compensation in the overlapping area. A fusion window is generated within the target seam area, and the fusion weights corresponding to the local data blocks on both sides of the seam are determined based on the distance from the pixels within the fusion window to the center line of the seam. Based on the fusion weight, the local data blocks after deformation compensation are weighted and fused for rendering to generate a finely stitched display image corresponding to the current display viewport.

10. The progressive loading method for multi-resolution pathological images according to claim 1, characterized in that, After generating the finely stitched display image corresponding to the current display viewport, the process further includes: When the current display viewport moves or scales, the changed display viewport position and the changed scaling factor are obtained, and the changed display viewport is mapped to the unified slice coordinate system corresponding to the low-resolution pyramid base map; Determine whether the changed display viewport covers the seam area where the fine-tuning transformation parameters have been determined. If it does, reuse the corresponding fine-tuning transformation parameters to perform deformation compensation and blending rendering on the local data blocks within the changed display viewport. If the changed display viewport covers the new seam area, then the new seam area is determined as the new target seam area, and the local data block clipping, end-side initial alignment, fine-tuning transformation parameter determination, and blending rendering are re-executed.

Images