A liver image visualization method and system based on three-dimensional reconstruction

Abstract

The application discloses a liver image visualization method and system based on three-dimensional reconstruction, which comprises the following steps: acquiring original liver image data to construct an initial three-dimensional model; accurately positioning and labeling lesions and blood vessels of the model, detecting coordinate deviation and real-time calibration to obtain a positioning model; judging defects of the model and adopting defect repair to obtain a complete model; extracting tissue density characteristics to adaptively stratify to obtain an optimized model; applying densification technology to interpolate and restore small blood vessels to obtain a dense model; if the data volume is over limit, compression encryption is carried out to obtain a transmission data packet; and the transmission data packet is remotely transmitted to a medical center to be decompressed and loaded to obtain a visualization result. The application realizes safe and efficient transmission of large-scale three-dimensional model data and remote consultation visualization support, so that the accuracy of liver lesion diagnosis of doctors and the remote medical cooperation efficiency are comprehensively improved.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing, and in particular relates to a method and system for visualizing liver images based on three-dimensional reconstruction. Background Technology

[0002] Three-dimensional visualization of liver images has become a key pillar of modern medical diagnosis and surgical planning. Because the liver has a complex structure involving numerous blood vessels and tissues, accurately reconstructing its spatial relationships is directly related to the reliability of subsequent disease diagnosis and the safety of treatment. It can provide doctors with intuitive observation of internal structures, help identify the location of lesions and their surrounding relationships, and thus support precise intervention.

[0003] Many current methods for processing liver images often struggle to effectively distinguish between different tissue densities, resulting in unclear representations of small blood vessels and tissue boundaries in the reconstructed model. This imperfection in thresholding can lead to the loss of model details, particularly in the blurring or incomplete reconstruction of microstructures, further amplifying localization errors. For example, during surgical planning, if the model cannot clearly display the distribution of small blood vessels, surgeons will find it difficult to accurately pinpoint the relative positions of lesions and critical structures, leading to errors in spatial coordinate calculations.

[0004] These issues are interconnected because insufficient threshold optimization directly affects the model's detail representation, while insufficient detail weakens the accurate localization calculation of internal structures, making it difficult for the reconstruction results to achieve the high resolution and completeness required in clinical applications. Especially in the presence of imaging defects or missing data, this inaccurate localization further exacerbates the model's instability.

[0005] Therefore, optimizing threshold processing during 3D reconstruction to improve model detail while repairing potential defects to ensure positioning accuracy has become a key issue in ensuring the reliability of liver image visualization and its clinical application value. Summary of the Invention

[0006] To address the above deficiencies, this invention provides a liver image visualization method and system based on 3D reconstruction. Through adaptive threshold optimization and densification techniques, it significantly improves the model's detail and resolution, making small blood vessels and tissue boundaries clearly discernible. Combined with a defect repair algorithm and a real-time coordinate calibration mechanism, it ensures high-precision localization of lesion areas and blood vessel distribution, enhancing the model's integrity and stability. Furthermore, by introducing compression encryption and remote transmission technologies, it achieves secure and efficient transmission of large-scale 3D model data and supports visualization for remote consultations. Specifically, the contents of this invention are as follows.

[0007] A liver image visualization method based on 3D reconstruction includes:

[0008] The original liver image data is acquired, and an initial three-dimensional model is constructed using a three-dimensional reconstruction algorithm. The initial three-dimensional model is then normalized in terms of spatial coordinates and segmented in terms of contour to obtain a preliminary reconstruction model containing the basic structure of the liver.

[0009] Based on the preliminary reconstruction model, the lesion area and blood vessel distribution are automatically labeled, coordinate deviations are detected in real time and calibrated using the Kalman filter algorithm to generate a precisely labeled localization model.

[0010] The positioning model is subjected to defect scanning, and a defect repair algorithm is used to perform pixel-level filling and edge optimization on the data missing areas to obtain a defect-free complete model.

[0011] Tissue density features are extracted from the complete model, and density stratification is achieved through an adaptive threshold algorithm to obtain a stratified optimized model.

[0012] The densification technique and interpolation calculation are applied to the hierarchical optimization model to restore the details of small blood vessels and tissue boundaries, resulting in a high-detail dense model.

[0013] If the data volume of the highly detailed dense model exceeds the transmission limit, gzip compression and an AES encryption layer are added to generate a transmission data packet.

[0014] The data packet is remotely sent to the medical center, and after decompression, visualization loading, and collaborative display, a three-dimensional visualization result of the liver supporting remote consultation is obtained.

[0015] Furthermore, the steps for constructing the preliminary reconstruction model specifically include:

[0016] The original image data is processed using a surface rendering algorithm to generate an initial 3D model.

[0017] The initial 3D model is normalized to a unified coordinate system to determine the connected regions of the liver contour and remove noise.

[0018] The liver parenchyma region was segmented using a region growing algorithm, and the normal vector of the model surface was smoothed to ensure consistency.

[0019] The inlet and outlet of blood vessels are located in the region of curvature change of the detection model, the center line of the blood vessels is extracted and the branch information is calculated to obtain a complete preliminary reconstruction model.

[0020] Furthermore, the step of generating the precisely labeled localization model specifically includes:

[0021] Preliminary annotation results of lesions and blood vessels were obtained through image segmentation;

[0022] The initial annotation results are subjected to secondary positioning and coordinate deviation detection. When the deviation exceeds the threshold, Kalman filter real-time calibration is triggered.

[0023] Key points were extracted and the annotation boundaries were optimized based on the continuity of blood vessels. Data consistency verification was performed to complete the localization model construction.

[0024] Furthermore, the steps to obtain the complete model specifically include:

[0025] The scanning and localization model identifies image defects and areas with missing data.

[0026] Perform pixel-level filling and optimize edge transitions for missing areas;

[0027] Iteratively perform defect detection and repair until the model is defect-free and a complete model is obtained.

[0028] Furthermore, the steps to obtain the hierarchical optimization model specifically include:

[0029] The tissue density distribution of the complete statistical model is used to divide high-density and low-density areas using an adaptive threshold.

[0030] Merge high- and low-density regions to generate a contrasting layered image;

[0031] Extract lesion contour information and determine the hierarchical optimization model structure.

[0032] Furthermore, the steps for obtaining the highly detailed and dense model specifically include:

[0033] Extracting fine structural information based on a hierarchical optimization model to identify low-resolution regions;

[0034] Interpolation calculations were used to restore the details of small blood vessels and tissue boundaries;

[0035] By verifying the integrity of details through a convolutional neural network and performing secondary interpolation on the unrecovered structure, a highly detailed and dense model is generated.

[0036] Furthermore, the step of generating the transmission data packet specifically includes:

[0037] Compare the data volume and transmission limitation threshold of high-detail dense models;

[0038] When the threshold is exceeded, gzip lossless compression is used, and an AES-256 encryption layer is added after compression;

[0039] Verify the integrity of the data packets to determine the final transmitted data packets.

[0040] Furthermore, the specific steps to obtain the visualization results of cross-regional consultations include:

[0041] Remotely transmit encrypted data packets to the medical center and then decompress and restore them;

[0042] Load the decompressed data into the visualization interface to complete the graphical mapping;

[0043] Data is categorized and organized according to consultation standards, and shared and feedback is collected through a medical collaboration platform;

[0044] The support vector machine algorithm is used to prioritize the displayed data and highlight key areas to complete the visualization of the consultation.

[0045] A liver image visualization system based on three-dimensional reconstruction, comprising:

[0046] The data acquisition and initial reconstruction module is used to acquire raw liver images and build a preliminary reconstruction model.

[0047] The precise positioning and calibration module is used to mark lesions and blood vessels and calibrate coordinate deviations in real time to obtain a positioning model.

[0048] The defect detection and repair module is used to scan for model defects and perform defect repair to obtain a complete model.

[0049] The feature extraction and hierarchical optimization module is used to extract density features and adaptively hierarchically obtain a hierarchical optimization model.

[0050] The detail enhancement and densification module is used to improve resolution and restore fine structures to obtain a highly detailed and dense model.

[0051] The data compression and encryption module is used to compress and encrypt model data to generate transmission data packets.

[0052] The remote transmission and visualization module is used for remote transmission, decompression and loading of data packets, and visualization of remote consultations.

[0053] Furthermore, the precise positioning and calibration module is configured with a Kalman filter algorithm to perform dynamic coordinate calibration; the data compression and encryption module uses a combination of gzip compression and AES encryption; and the remote transmission and visualization module supports data priority display based on support vector machines.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] (i) Enhancing the accuracy and anatomical realism of 3D reconstruction to achieve high-precision localization of lesions and blood vessels: A combined reconstruction process of surface rendering, region growth with additive vector smoothing, and extraction of blood vessel centerlines is adopted to achieve standardized coordinate unification of liver contour, parenchyma, and vascular structure, significantly improving the geometric consistency and anatomical realism of the model. Then, through secondary fine localization and real-time coordinate calibration with Kalman filtering, the annotation deviation is greatly reduced, ensuring the accurate and reliable spatial location of lesion areas and blood vessel distribution, providing stable support for accurate diagnosis and surgical planning.

[0056] (ii) Automatically repairs model defects, resulting in greater integrity; the built-in defect scanning and defect iteration repair mechanism can automatically fill in missing data, optimize boundary transitions, eliminate model distortion caused by image noise and acquisition interference, and output a defect-free and complete 3D model.

[0057] (III) Adaptive density stratification enhances tissue recognition; enhanced densification provides high-definition representation of fine structures. Adaptive threshold stratification based on tissue density features clearly distinguishes between high and low density areas, highlighting lesion outlines and vascular boundaries, allowing doctors to intuitively identify the relationship between lesions and surrounding tissues. Through interpolation calculation, CNN feature extraction, and secondary detail repair, fine blood vessels and tissue boundaries are restored, significantly improving model resolution and detail representation, solving the problem of detail loss in traditional reconstruction.

[0058] (iv) Secure and efficient transmission, adapted to telemedicine; supports remote consultation and collaborative diagnosis; employs gzip lossless compression and AES encryption to reduce data volume without losing detail, balancing transmission efficiency and medical data security, and meeting the needs of large-scale 3D model remote transmission. Visual interface loading, collaborative data priority sorting, and dynamic presentation of key areas enable efficient cross-regional consultations, comprehensively improving the accuracy of liver disease diagnosis and the efficiency of telemedicine collaboration; thereby comprehensively improving the accuracy of doctors' diagnosis of liver lesions and the efficiency of telemedicine collaboration. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowcharts, in some cases the steps shown or described may be executed in a different order than that shown here.

[0062] Example 1: As Figure 1 As shown, this embodiment provides a liver image visualization method based on three-dimensional reconstruction, including:

[0063] Original liver imaging data is acquired, an initial three-dimensional model is constructed using a three-dimensional reconstruction algorithm, and the spatial coordinates in the initial three-dimensional model are preliminarily calculated to obtain a preliminary reconstruction model containing basic structural information.

[0064] Based on the preliminary reconstruction model, the lesion area and blood vessel distribution are automatically marked using precise positioning technology. If the coordinate deviation is detected to exceed the preset threshold during the marking process, the coordinate parameters are adjusted through a real-time calibration algorithm to determine the precise marking positioning model.

[0065] For the precisely labeled positioning model, determine whether there are image defects or missing data. If so, use a defect repair algorithm to fill in the missing parts and optimize the boundaries to obtain a complete repaired model.

[0066] Tissue density features are extracted from the repaired complete model, and a threshold optimization algorithm is used to adaptively stratify the density values ​​to determine the optimized model after stratification.

[0067] Based on the optimized model after layering, densification technology is applied to improve resolution and detail representation. By interpolation calculation, small blood vessels and tissue boundaries are restored to obtain a dense model with high detail representation.

[0068] Obtain the data volume of the high-detail dense model, determine whether the data volume exceeds the transmission limit, and if it does, use a compression algorithm to reduce the data volume and add an encryption layer to determine the compressed and encrypted transmission data packet.

[0069] The compressed and encrypted data packets are sent to the telemedicine center using long-distance transmission technology. The data packets are then decompressed and a visualization interface is loaded at the telemedicine center to obtain visualization results that support remote consultations.

[0070] Furthermore, the process of obtaining a preliminary reconstruction model containing basic structural information includes:

[0071] Raw liver imaging data was acquired, and a surface rendering 3D reconstruction algorithm was used to process the raw imaging data to obtain an initial 3D model.

[0072] The spatial coordinates of all vertices in the initial 3D model are normalized to a unified coordinate system to obtain the initial 3D model with standardized coordinates.

[0073] Based on the initial 3D model with standardized coordinates, the adjacency relationship between each vertex is calculated, the connected region of the main contour surface of the liver is determined, and a preliminary reconstruction model after contour segmentation is obtained.

[0074] For the preliminary reconstructed model after contour segmentation, the region growing algorithm is used to expand from the high-density region inside the liver. If the difference between the current voxel density and the seed point density is within a preset threshold range, it is included in the same region, thus obtaining an internal structure model that includes the liver parenchyma region.

[0075] Based on the internal structure model, the consistency of the surface normal vector direction is calculated. If the angle between the normal vectors of adjacent facets is less than a preset threshold, the smooth connection is retained, thus obtaining the optimized reconstruction model with surface smoothing.

[0076] The optimized reconstruction model with smoothed surface was used to detect coordinate regions with drastic curvature changes, determine the location of tubular structures at the inlet and outlet of liver vessels, and obtain a fine reconstruction model with annotations of the main vascular structures.

[0077] Based on the finely reconstructed model annotated with the main vascular structures, the coordinate sequence of the vascular centerline is extracted, and the changes in the centerline curvature and diameter are calculated to obtain a complete three-dimensional liver model with vascular path and branch information.

[0078] Furthermore, in this embodiment, when acquiring raw liver image data, medical imaging equipment such as CT or MRI scans are used to obtain a series of two-dimensional slice images. These images contain density information and spatial location data of the liver, laying the foundation for subsequent three-dimensional reconstruction. Assuming the scan yields 100 slices, each with a resolution of 512×512 pixels and a slice thickness of 1 mm, a preliminary raw dataset containing liver volume information can be constructed.

[0079] In one possible implementation, this embodiment uses a surface rendering 3D reconstruction algorithm such as MarchingCubes to process the raw image data, extracting isosurfaces based on voxel density values ​​to generate an initial 3D model. Assuming a density value of 120 is used as a threshold to extract the liver surface, the generated model contains approximately 100,000 vertices and 200,000 triangular facets, initially presenting the external morphology of the liver.

[0080] For example, this embodiment performs a unified coordinate system normalization process for vertex spatial coordinates, mapping the coordinates of all vertices to the range of 0 to 1. Assuming the original coordinate range is -200 to 200 mm for the x-axis, -150 to 150 mm for the y-axis, and -100 to 100 mm for the z-axis, the coordinates are standardized through linear transformation, which facilitates subsequent calculations and consistency analysis.

[0081] In one possible implementation, this embodiment calculates vertex adjacency relationships by constructing a triangular patch adjacency graph to determine the connected regions of the main contour surfaces of the liver. Assuming two main connected regions are detected, one covering the main body of the liver and the other being a noisy, small region, the small region can be removed to obtain a preliminary reconstructed model after contour segmentation.

[0082] For example, in this embodiment, when using a region growing algorithm, a high-density region inside the liver, such as a voxel with a density value of 150, is used as a seed point. A threshold range of ±20 is set, and voxels of similar density are gradually expanded and incorporated to ultimately form an internal structural model containing the liver parenchyma. This helps to distinguish liver tissue from surrounding low-density fat or cavities.

[0083] In one possible implementation, when calculating the consistency of surface normal vector directions in this embodiment, if the angle between the normal vectors of adjacent facets is less than 30 degrees, then a smooth connection is retained. Assuming a region has 1000 facets, of which 900 meet the condition, the smoothed surface appears more natural, improving the model's visual effect and the accuracy of subsequent analysis.

[0084] For example, in this embodiment, when detecting regions with drastic curvature changes, the tubular structures at the inlets and outlets of hepatic vessels are located. Assuming that the rate of curvature change exceeds 0.5 in the porta hepatis region, it is labeled as a vessel inlet based on morphological features, and a fine reconstruction model is generated.

[0085] In one possible implementation, this embodiment extracts the coordinate sequence of the blood vessel centerline and calculates the changes in curvature and diameter. For example, if the curvature of a certain blood vessel increases from 0.1 to 0.3 and the diameter decreases from 5 mm to 2 mm, it can be identified as a branch point, and a complete three-dimensional liver model can be constructed. This not only improves the model detail but also provides data support for hemodynamic analysis.

[0086] Furthermore, the process of determining the precisely labeled localization model includes:

[0087] Data processing is performed based on the preliminary reconstruction model to obtain the initial annotation information of the lesion area and blood vessel distribution. Image segmentation technology is used to separate the target area to obtain the preliminary annotation results.

[0088] The preliminary annotation results are used to perform secondary localization processing on the lesion area and blood vessel distribution using precise positioning technology, to detect whether there is coordinate deviation, and to determine the deviation detection data.

[0089] If the deviation detection data exceeds the preset threshold, a real-time calibration mechanism is triggered, and the coordinate parameters are dynamically adjusted using the Kalman filter algorithm to obtain the calibrated coordinate information.

[0090] Based on the calibrated coordinate information, update the labeled positions of the lesion area and blood vessel distribution to generate intermediate results with accurate labels;

[0091] Key point data are extracted from the intermediate results of precise annotation, and the annotation boundary is optimized by combining the continuity characteristics of blood vessel distribution to obtain annotation details.

[0092] Based on the annotation details, a complete localization model is constructed, integrating precise information on the lesion area and blood vessel distribution to obtain a stable model output;

[0093] Perform data consistency verification on the model output to confirm the matching degree between all annotation information and the preliminary reconstruction model, and complete the final construction of the localization model.

[0094] Furthermore, in this embodiment, when annotating the lesion area and blood vessel distribution based on the three-dimensional reconstruction model of the liver, initial annotation information is first obtained through data processing of the preliminary reconstruction model. Specifically, threshold segmentation is performed on the voxel data of the preliminary reconstruction model to separate lesion areas with abnormal density and high-contrast blood vessel structures, thereby obtaining preliminary annotation results.

[0095] Specifically, in this embodiment, the liver parenchyma density range can be set to 50-150 HU. Areas outside this range are initially marked as potential lesions, while tubular high-density structures are marked as vascular distribution.

[0096] In one possible implementation, the preliminary annotation results are processed using precise positioning technology for secondary positioning, mainly to detect coordinate deviations.

[0097] For example, by comparing the matching degree between the labeled boundary and the surface normal vector of the reconstructed model, the offset distance of the boundary points is calculated. If the average offset exceeds 2mm, a deviation is identified, and deviation detection data is generated. This deviation often originates from image noise or surface fluctuations in the reconstruction algorithm and needs to be corrected in a timely manner.

[0098] It should be noted that if the deviation detection data exceeds a preset threshold, such as a threshold of 3 mm, a real-time calibration mechanism is triggered. At this time, the Kalman filter algorithm is used to dynamically adjust the coordinate parameters. The Kalman filter, through prediction and measurement update cycles, gradually fuses the current coordinate observations with historical state estimates, significantly reducing the impact of noise and obtaining calibrated coordinate information. In the liver model, this calibration can reduce the deviation of the blood vessel inlet position from the initial 4.5 mm to within 1.2 mm, improving positioning stability.

[0099] For example, in this embodiment, after updating the labeled positions of the lesion area and blood vessel distribution based on the calibrated coordinate information, an intermediate result with accurate labeling is generated. This result preserves the complete outline of the lesion and the continuous path of the blood vessels, avoiding the breaks or offsets commonly found in the initial labeling. Key point data, such as the lesion center point and blood vessel branch nodes, are extracted from the intermediate result with accurate labeling, and the labeling boundary is optimized by combining the continuity characteristics of the blood vessel distribution.

[0100] Specifically, this embodiment examines the connectivity of adjacent voxels by extending along the centerline of the blood vessel, smooths the boundary transition, and determines the final annotation details, thereby making the lesion edge more consistent with the actual anatomical structure.

[0101] In one embodiment, this embodiment constructs a complete localization model through final annotation details, integrating precise information on the lesion area and vascular distribution into a stable model output. This output not only preserves the main structures of the liver but also clearly shows the spatial relationship between the lesion location and the blood vessels, facilitating subsequent surgical planning by the surgeon. Data consistency verification is performed on the stable model output, such as comparing the voxel overlap rate between the annotation information and the preliminary reconstructed model layer by layer. An overlap rate of over 95% is required to confirm the matching degree and complete the final construction of the localization model. This verification ensures the reliability of the annotations, avoids overall model distortion caused by local deviations, and ultimately provides a highly reliable three-dimensional reference for precise liver surgery navigation, improving the accuracy and safety of clinical decision-making.

[0102] Furthermore, the process of obtaining a fully repaired model includes:

[0103] Perform a defect scan on the positioning model to determine if there are any image defects or missing data;

[0104] If the judgment result indicates that a defect exists, then the defect location and defect area range are extracted;

[0105] A defect repair algorithm is used to fill in the missing areas at the pixel level, generating a preliminary repair model;

[0106] Based on the consistency between the filled area and the surrounding pixels in the preliminary repair model, the edge transition is optimized to obtain the edge optimization model;

[0107] Based on the edge optimization model, a defect scan is performed again to determine whether there are still residual defects.

[0108] If residual defects still exist, repeat the defect repair algorithm and edge optimization until the defect scan result is defect-free, and obtain the complete positioning model after repair.

[0109] Furthermore, after obtaining the precisely labeled positioning model, this embodiment performs a defect scan to determine whether there are image defects or missing data. This process can be understood as a comprehensive check of the model's output quality, aiming to ensure the model's reliability in subsequent applications. Suppose that in a scan, the system detects that some pixels in the image data of a lesion area are missing. This missing data may be due to signal interference during the acquisition process of the imaging device. To address this issue, the system records the specific location of the missing area, for example, a rectangular area in the image coordinate system ranging from 100 to 120 on the horizontal axis and 50 to 70 on the vertical axis.

[0110] Specifically, this embodiment employs a defect repair algorithm to fill in the extracted defect locations and missing area ranges at the pixel level. One possible implementation is to estimate the pixel values ​​of the missing area using interpolation methods based on the color and texture features of surrounding pixels.

[0111] For example, pixels surrounding the missing area exhibit a relatively smooth grayscale transition. The system generates filling pixels based on this characteristic, making the repaired area visually consistent with its surroundings. After the initial repair model is generated, subtle unnatural transitions may be observed between the filled area and surrounding pixels at the edges. Further optimization of the edge transition is then necessary. Assuming that during optimization, the system adjusts the grayscale value range of edge pixels using smoothing filtering techniques, narrowing the original abrupt change range to a more natural gradient range.

[0112] For example, in this embodiment, after obtaining the edge-optimized model, a defect scan is performed again to confirm whether there are still residual defects. Suppose the second scan finds small areas of inconsistent color, such as within a tiny area between 110 and 112 on the horizontal axis and 55 and 57 on the vertical axis, the system will repeatedly execute the defect repair algorithm and edge optimization until the defect scan results show no abnormalities. This iterative process ensures that the final output complete localization model achieves a high standard of image quality.

[0113] It should be noted that for localization models of lesion areas and vascular distribution, the steps of repairing defects and optimizing edges are particularly important, because subtle differences in these areas may affect the accuracy of subsequent analyses.

[0114] Specifically, throughout the entire repair process, each step of this embodiment focuses on the precise annotation of the lesion area and blood vessel distribution. For example, defect scanning not only addresses missing pixels but also checks whether the annotation boundaries are broken due to missing data. If a blood vessel distribution annotation is found to be interrupted along a path during a scan, the system will combine the continuity characteristics of the blood vessels and use path prediction technology to complete the interrupted portion. This approach effectively improves the integrity of the model while ensuring the coherence of the annotation information, providing a reliable data foundation for subsequent applications. The beneficial effect of this processing method is that both the lesion area and the blood vessel distribution can be output in a more complete and natural way, avoiding the risk of misjudgment due to defects.

[0115] For example, once a complete localization model is generated, its high consistency and defect-free characteristics in image presentation provide stable support for subsequent analysis. Assuming that in one application, the complete model clearly shows the boundary of the lesion area and the direction of blood vessel distribution, this provides an intuitive and reliable reference for subsequent medical diagnosis. The entire process, through multiple safeguards of defect scanning, repair, and optimization, ensures high-fidelity model performance in detail, significantly enhancing the practical value of the data output.

[0116] Furthermore, the process of determining the optimization model after hierarchical processing includes:

[0117] Tissue density feature data are obtained from the complete model, and the stratification boundary is determined by an adaptive threshold algorithm based on the data distribution of density values.

[0118] If the density value is greater than the layer boundary, it is marked as a high-density area;

[0119] If the density value is less than or equal to the stratification boundary, it is marked as a low-density area;

[0120] Merge high-density and low-density areas to generate a layered image with contrasting differences;

[0121] Identify and extract contour information of lesion areas from layered images;

[0122] Based on the contour information, the structure of the final optimized hierarchical model is determined.

[0123] Furthermore, in this embodiment, when acquiring tissue density feature data in the complete repair model, a comprehensive scan of the image data is first performed to extract the density value information of each pixel. This process aims to capture the density distribution characteristics of tissue in different regions, providing a data foundation for subsequent stratification. For example, in a certain scan, the system records a density value range of 50 to 200 for a certain region, with the density values ​​of most pixels concentrated in the range of 80 to 120, showing obvious distribution characteristics.

[0124] For example, in this embodiment, when using an adaptive threshold algorithm to determine the layer boundaries for density value data distribution, the threshold can be dynamically adjusted based on the statistical characteristics of the data distribution. Suppose analysis reveals that the density value distribution exhibits a bimodal characteristic, the system will automatically select the median value between the two peaks, such as 100, as the layer boundary. This method can adapt to the characteristics of different image data, ensuring the rationality of the layer boundaries.

[0125] For example, when labeling high-density and low-density areas, if a region's density value is greater than 100, it is labeled as a high-density area and displayed as a brighter image region; if the density value is less than or equal to 100, it is labeled as a low-density area and appears as a darker area. Assuming that in an image, high-density areas are mainly concentrated in the central region, which may be areas with denser tissue structures, while low-density areas are distributed on the periphery, which may be areas with looser tissue structures.

[0126] For example, when merging high-density and low-density areas to generate a layered image with contrast, this embodiment highlights the difference by enhancing the grayscale contrast between the two regions. Assuming that in the generated image, high-density areas are displayed in white and low-density areas in dark gray, the boundaries between the two regions are clearly visible, facilitating subsequent analysis. This method of presenting contrast helps to more intuitively observe the distribution of tissue structures.

[0127] For example, in this embodiment, when identifying and extracting the contour information of a lesion region from a layered image, edge detection technology is used to lock the boundary of the lesion region. Suppose that in a certain detection, the system identifies an irregularly shaped lesion region whose contour line ranges from 150 to 180 on the horizontal axis and from 90 to 110 on the vertical axis in the image coordinate system. By extracting this contour information, accurate boundary data can be provided for subsequent modeling.

[0128] For example, in this embodiment, when determining the final optimized hierarchical model structure based on the extracted contour information, a comprehensive adjustment is made by combining contour shape and density distribution features. Suppose that during the optimization process, the system finds that the contour boundary of the lesion area highly overlaps with the high-density area, it may further refine the boundary lines to make them more closely resemble the actual tissue structure. This adjustment can improve the model's expressiveness in detail, providing a more reliable reference for subsequent analysis. For example...

[0129] It should be noted that, by combining layered images and contour information, this embodiment can clearly present the density difference between the lesion area and the surrounding tissue. This differentiated presentation provides important clues for subsequent judgment and significantly improves the accuracy of the analysis.

[0130] Furthermore, the process of obtaining a dense model with high detail includes:

[0131] Based on the hierarchical optimization model data, a pre-established image processing framework is used to obtain the fine structural information in the original image, resulting in a preliminary hierarchical data set.

[0132] By analyzing the preliminary layered dataset, a compaction technique is applied to address the resolution enhancement requirements and determine the distribution of low-resolution regions in the image.

[0133] Based on the distribution of low-resolution regions, interpolation calculation methods are implemented to restore the detailed information of small blood vessels and tissue boundaries, and to obtain an enhanced detailed dataset.

[0134] For the enhanced detail dataset, a convolutional neural network algorithm is used to extract high-detail features, determine whether there are unrecovered fine structures, and if there are unrecovered structures, perform secondary interpolation to obtain a complete detailed image.

[0135] By further processing the complete detailed image, a dense model is constructed, the boundary restoration effect of each region in the model is obtained, and the final boundary sharpness is determined.

[0136] Based on the final boundary clarity, the application technology module integrates the dense model to obtain an output model with high detail.

[0137] Furthermore, this embodiment achieves high-precision restoration of blood vessel and tissue boundary details in medical images through layered optimized model data. For acquiring fine structural information from the original image, assuming a 512×512 resolution vascular CT image is being processed, a pre-built image processing framework is first used to extract the contour information of tiny blood vessels in the image through edge detection technology, forming a preliminary layered data set. This process effectively separates background noise from the target structure, ensuring the accuracy of subsequent analysis.

[0138] For example, in this embodiment, when analyzing the initial layered dataset, regions with insufficient resolution are identified. Assuming that the resolution of some blood vessel regions in the image is only 1 / 4 of the original image, a densification technique is applied using a pixel density enhancement algorithm to visualize the distribution of low-resolution regions, clearly identifying the areas requiring focused processing. This method helps to accurately locate problem areas, providing a basis for subsequent detail restoration.

[0139] For example, in this embodiment, when implementing interpolation calculations for low-resolution regions, bilinear interpolation is used to restore detailed information about small blood vessels and tissue boundaries. Assuming that the pixel value of a certain blood vessel edge is improved from a blurry grayscale value of 50 to a clear grayscale value of 80 after interpolation, the enhancement effect on the detail dataset is significant. This method can effectively improve the local sharpness of the image, laying the foundation for subsequent feature extraction.

[0140] For example, in this embodiment, when extracting high-detail features from the enhanced detail dataset using a convolutional neural network algorithm, multiple convolutional operations are used to determine whether there are unrecovered fine structures. If the network identifies a region with blurred boundaries, it performs secondary interpolation until the detailed image is complete. This deep learning method can intelligently identify problem areas and improve the overall image quality.

[0141] For example, in this embodiment, when constructing the dense model, the complete detailed image is further processed to analyze the boundary restoration effect of each region. Assuming the boundary sharpness of a certain tissue is improved from an initial 60% to 90%, the final boundary sharpness is confirmed. This model construction method can optimize image performance and ensure the accuracy of detail rendering.

[0142] For example, in this embodiment, when integrating the dense model using the application technology module to achieve the final boundary sharpness result, multi-scale fusion technology is employed to uniformly output high-detail features from different regions. Assuming the output model improves vascular detail by a factor of two compared to the original image, its high-detail performance provides doctors with a more reliable basis for diagnosis. This integration method significantly enhances the model's practical value. Through the above multi-faceted technical implementations and examples, it can be seen that each step is closely focused on image detail restoration, progressing step by step to ensure the best application effect of the final output model in the field of medical imaging, while providing strong technical support for subsequent doctor diagnosis and research.

[0143] Furthermore, the process of determining the compressed and encrypted transmission data packets includes:

[0144] Obtain the data volume of the high-detail dense model, compare the data volume with the preset transmission limit, and determine whether the data volume exceeds the transmission limit.

[0145] If the data size exceeds the transmission limit, the gzip compression algorithm is used to compress the data size to obtain the compressed data size.

[0146] AES encryption is added based on the compressed data volume to obtain encrypted data packets;

[0147] Verify the integrity of the encrypted data packet to determine the final transmitted data packet.

[0148] Furthermore, in this embodiment, when obtaining the data volume of a highly detailed and dense model, the overall volume of the model constructed from the complete detailed image is first statistically analyzed.

[0149] Specifically, this embodiment calculates the storage footprint of all vertices, faces, and associated high-resolution detail data by traversing the model's mesh data and texture layers, thus obtaining an accurate volume value.

[0150] In one embodiment, it is assumed that the original high-detail dense model contains approximately 5 million vertices and corresponding high-resolution texture maps, and its data volume may reach 850MB.

[0151] Understandably, the data volume will then be compared with a preset transmission limit. In this embodiment, the medical image transmission system sets a single transmission limit of 500MB to accommodate network bandwidth and storage constraints. If the data volume does not exceed the transmission limit, for example, if it is 420MB, the original high-detail dense model is directly packaged into a transmission data packet. This preserves the complete details of all small blood vessels and tissue boundaries, avoiding any potential information loss, thereby ensuring that the receiving end can directly use it for subsequent 3D reconstruction or diagnostic analysis.

[0152] It should be noted that when the data size exceeds the transmission limit, for example, reaching 850MB, the gzip compression algorithm will be used first. gzip achieves efficient lossless compression by identifying repeating patterns in the model data, such as similar coordinates of adjacent meshes or redundant pixels in textures.

[0153] In one possible implementation, applying gzip can significantly reduce the data size from 850MB to approximately 320MB. This compression not only greatly reduces transmission time but also preserves the integrity of all high-detail features of the model, facilitating rapid decompression and restoration on a medical workstation.

[0154] Specifically, this embodiment adds an AES encryption layer immediately after compression, using the AES-256 standard to encrypt the compressed data packet, generating a secure encrypted data packet. This encryption layer ensures that sensitive blood vessel and tissue details are protected from unauthorized access during transmission, meeting medical data privacy requirements. At the same time, the encryption process does not significantly increase data volume, adding only a small amount of overhead, effectively balancing security and transmission efficiency.

[0155] For example, after encryption, this embodiment verifies the integrity of the encrypted data packet by calculating a checksum or hash value. If the verification passes, the packet is determined to be the final transmission data packet and can be directly used for network transmission. This integrity verification can promptly detect any damage before transmission, ensuring that the receiving end receives data that is completely consistent with the original high-detail dense model, thereby supporting accurate visualization of fine structures and clinical assessment, and significantly improving the reliability and efficiency of the overall workflow.

[0156] Furthermore, the process of obtaining visual results that support remote consultations includes:

[0157] Using long-distance transmission technology, compressed and encrypted data packets are sent from the local system to the telemedicine center, thus obtaining the data packets transmitted to the telemedicine center;

[0158] The received data packets are decompressed at the telemedicine center to restore the data content and obtain the decompressed medical data.

[0159] Based on the decompressed medical data, it is loaded into a pre-established visualization interface, and the data content is transformed into a graphical display through data mapping technology to obtain preliminary visualization results.

[0160] Based on the preliminary visualization results and the needs for supporting remote consultations, the data is categorized and organized. If the categorized data meets the preset consultation criteria, it is determined to be visualization content that can be used for consultations.

[0161] For visual content that can be used for consultation, data is shared through a medical collaboration platform to obtain feedback from remote participants and obtain shared collaborative data.

[0162] Based on the shared collaborative data, the support vector machine algorithm is used to prioritize the data and determine the data that needs to be highlighted. Through remote data loading technology, the data that needs to be highlighted is visualized to obtain a visualization result that supports remote consultation.

[0163] Furthermore, in this embodiment, when sending the compressed and encrypted data packets from the local system to the remote medical center via long-distance transmission technology, a high-speed network protocol is used to ensure the stability of data transmission.

[0164] Specifically, assuming the local system is located in a city hospital and the telemedicine center is located in a provincial medical hub hundreds of kilometers away, the data packet size is 320MB, and a dedicated fiber optic network is used during transmission to ensure that the data is initially transmitted within 5 minutes.

[0165] It should be noted that after the transmission is completed, the telemedicine center will record the data packet reception time and integrity status, forming a transmission log for subsequent verification.

[0166] For example, in this embodiment, when decompressing the received data packets at the telemedicine center, the preset decompression tool can be a gzip-based reverse decompression software with a simple interface, suitable for medical personnel without a technical background.

[0167] In one embodiment, the decompression process runs automatically on a dedicated server, and a 320MB data packet can be restored to 850MB of original medical data within 2 minutes. At the same time, a decompression report is generated, recording whether there is any data corruption.

[0168] For example, when loading decompressed medical data into a visualization interface, data mapping technology can transform complex mesh data and texture information into 3D images. Suppose the data contains a highly detailed vascular model; the system will automatically map vertex and facet data into an intuitive 3D graphic, allowing doctors to rotate or zoom to view details.

[0169] It should be noted that the initial visualization results are stored as temporary files for easy adjustment later.

[0170] For example, when classifying and organizing data based on preliminary visualization results and the needs of remote consultations, the data is divided into two categories: key areas and non-key areas. Assuming the consultation criteria require that the resolution of key area data be no less than 90% of the original data, only data meeting this standard will be identified as usable for consultation. This classification method ensures the clear presentation of key information during consultations.

[0171] For example, when sharing visualizations through a medical collaboration platform, remote participants may be multiple experts who view the data in real time and submit feedback through the platform.

[0172] In one embodiment, the platform supports text and voice feedback, allowing experts to annotate specific areas on images and create collaborative data. This sharing mechanism significantly improves communication efficiency.

[0173] For example, when prioritizing shared collaborative data using a support vector machine algorithm, the system determines which data needs to be highlighted based on feedback frequency and annotation importance. If a vascular region is highlighted as key by multiple experts, the system will increase its priority, and the final visualization model will prominently display that region.

[0174] It should be noted that this sorting method ensures that the focus of the consultation remains consistent.

[0175] For example, when distributing the final visualization model to relevant terminals using remote data loading technology, cloud-based distribution services can be used to ensure synchronized updates. Assuming the terminals include large-screen devices in multiple consultation rooms, the model can be loaded within 10 minutes, forming a complete presentation supporting remote consultations. This distribution method ensures the timeliness of information access for all parties.

[0176] Example 2: As Figure 2 As shown, based on the same inventive concept, this embodiment also provides a liver image visualization system based on three-dimensional reconstruction, including:

[0177] The data acquisition and initial reconstruction module is used to acquire raw liver image data, construct an initial three-dimensional model through a three-dimensional reconstruction algorithm, perform preliminary calculations on the spatial coordinates in the initial three-dimensional model, and obtain a preliminary reconstruction model containing basic structural information.

[0178] The precise positioning and calibration module is used to automatically label the lesion area and blood vessel distribution based on the preliminary reconstruction model using precise positioning technology. If the coordinate deviation is detected to exceed the preset threshold during the labeling process, the coordinate parameters are adjusted through a real-time calibration algorithm to determine the precise positioning model.

[0179] The defect detection and repair module is used to determine whether there are image defects or missing data in the precisely annotated positioning model. If so, the defect repair algorithm is used to fill in the missing parts and optimize the boundaries to obtain a complete repaired model.

[0180] The feature extraction and hierarchical optimization module is used to extract tissue density features from the repaired complete model, and to use a threshold optimization algorithm to adaptively hierarchically process the density values ​​to determine the optimized model after hierarchical processing.

[0181] The detail enhancement and densification module is used to improve the resolution and detail performance of the optimized model after layering processing by applying densification technology. It restores small blood vessels and tissue boundaries through interpolation calculation to obtain a dense model with high detail performance.

[0182] The data compression and encryption module is used to obtain the data volume of the high-detail dense model, determine whether the data volume exceeds the transmission limit, and if it does, use a compression algorithm to reduce the data volume and add an encryption layer to determine the compressed and encrypted transmission data packet.

[0183] The remote transmission and visualization module is used to send compressed and encrypted data packets to the telemedicine center via long-distance transmission technology. The telemedicine center then decompresses the data packets and loads a visualization interface to obtain visualization results that support remote consultations.

[0184] The liver image visualization system based on three-dimensional reconstruction provided in this embodiment has all the advantages of the liver image visualization method based on three-dimensional reconstruction provided in Embodiment 1.

[0185] Example 3: This example also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Example 1.

[0186] Example 4: This example also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Example 1.

[0187] Example 5: This example also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Example 1.

[0188] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A liver image visualization method based on three-dimensional reconstruction, characterized in that, include: The original liver image data is acquired, and an initial three-dimensional model is constructed using a three-dimensional reconstruction algorithm. The initial three-dimensional model is then normalized in terms of spatial coordinates and segmented in terms of contour to obtain a preliminary reconstruction model containing the basic structure of the liver. Based on the preliminary reconstruction model, the lesion area and blood vessel distribution are automatically labeled, coordinate deviations are detected in real time and calibrated using the Kalman filter algorithm to generate a precisely labeled localization model. The positioning model is subjected to defect scanning, and a defect repair algorithm is used to perform pixel-level filling and edge optimization on the data missing areas to obtain a defect-free complete model. Tissue density features are extracted from the complete model, and density stratification is achieved through an adaptive threshold algorithm to obtain a stratified optimized model. The densification technique and interpolation calculation are applied to the hierarchical optimization model to restore the details of small blood vessels and tissue boundaries, resulting in a high-detail dense model. If the data volume of the highly detailed and dense model exceeds the transmission limit, gzip compression and an AES encryption layer are added to generate a transmission data packet. The data packet is remotely sent to the medical center, and after decompression, visualization loading, and collaborative display, a three-dimensional visualization result of the liver supporting remote consultation is obtained.

2. The liver image visualization method based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps for constructing the preliminary reconstruction model include: The original image data is processed using a surface rendering algorithm to generate an initial 3D model. The initial 3D model is normalized to a unified coordinate system to determine the connected regions of the liver contour and remove noise. The liver parenchyma region was segmented using a region growing algorithm, and the normal vector of the model surface was smoothed to ensure consistency. The inlet and outlet of blood vessels are located in the region of curvature change of the detection model, the center line of the blood vessels is extracted and the branch information is calculated to obtain a complete preliminary reconstruction model.

3. The liver image visualization method based on three-dimensional reconstruction according to claim 1, characterized in that, The steps for generating the precisely labeled localization model specifically include: Preliminary annotation results of lesions and blood vessels were obtained through image segmentation; The initial annotation results are subjected to secondary positioning and coordinate deviation detection. When the deviation exceeds the threshold, Kalman filter real-time calibration is triggered. Key points were extracted and the annotation boundaries were optimized based on the continuity of blood vessels. Data consistency verification was performed to complete the localization model construction.

4. The liver image visualization method based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps to obtain the complete model include: The scanning and localization model identifies image defects and areas with missing data. Perform pixel-level filling and optimize edge transitions for missing areas; Iteratively perform defect detection and repair until the model is defect-free and a complete model is obtained.

5. The liver image visualization method based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps to obtain the hierarchical optimization model include: The tissue density distribution of the complete statistical model is used to divide high-density and low-density areas using an adaptive threshold. Merge high- and low-density regions to generate a contrasting layered image; Extract lesion contour information and determine the hierarchical optimization model structure.

6. The liver image visualization method based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps to obtain the highly detailed and dense model include: Extracting fine structural information based on a hierarchical optimization model to identify low-resolution regions; Interpolation calculations were used to restore the details of small blood vessels and tissue boundaries; By verifying the integrity of details through a convolutional neural network and performing secondary interpolation on the unrecovered structure, a highly detailed and dense model is generated.

7. The liver image visualization method based on three-dimensional reconstruction according to claim 1, characterized in that, The steps for generating the transmission data packet specifically include: Compare the data volume and transmission limitation threshold of high-detail dense models; When the threshold is exceeded, gzip lossless compression is used, and an AES-256 encryption layer is added after compression; Verify the integrity of the data packets to determine the final transmitted data packets.

8. The liver image visualization method based on three-dimensional reconstruction according to claim 1, characterized in that, The specific steps to obtain visualized results of remote consultations include: Remotely transmit encrypted data packets to the medical center and then decompress and restore them; Load the decompressed data into the visualization interface to complete the graphical mapping; Data is categorized and organized according to consultation standards, and shared and feedback is collected through a medical collaboration platform; The support vector machine algorithm is used to prioritize the displayed data and highlight key areas to complete the visualization of the consultation.

9. A liver image visualization system based on three-dimensional reconstruction, characterized in that, include: The data acquisition and initial reconstruction module is used to acquire raw liver images and build a preliminary reconstruction model. The precise positioning and calibration module is used to mark lesions and blood vessels and calibrate coordinate deviations in real time to obtain a positioning model. The defect detection and repair module is used to scan for model defects and perform defect repair to obtain a complete model. The feature extraction and hierarchical optimization module is used to extract density features and adaptively hierarchically obtain a hierarchical optimization model. The detail enhancement and densification module is used to improve resolution and restore fine structures to obtain a highly detailed and dense model. The data compression and encryption module is used to compress and encrypt model data to generate transmission data packets. The remote transmission and visualization module is used for remote transmission, decompression and loading of data packets, and visualization of remote consultations.

10. A liver image visualization system based on three-dimensional reconstruction according to claim 9, characterized in that, The precise positioning and calibration module is configured with a Kalman filter algorithm to perform dynamic coordinate calibration; the data compression and encryption module uses a combination of gzip compression and AES encryption; and the remote transmission and visualization module supports data priority display based on support vector machine.

Images