Method for reconstructing pulmonary vascular tree based on multi-scale watershed segmentation

Abstract

The present invention relates to the field of medical image processing technology, and more specifically, to a method for reconstructing a pulmonary vascular tree based on multi-scale watershed segmentation, comprising: acquiring pulmonary CT image data; performing image preprocessing operations based on the pulmonary CT image data to obtain preprocessed image data; performing multi-scale analysis based on the preprocessed image data to obtain multi-scale feature data; performing a watershed segmentation algorithm based on the multi-scale feature data to obtain a vascular segmentation result; performing a vascular tree reconstruction operation based on the vascular segmentation result to obtain reconstructed pulmonary vascular tree data; and outputting the reconstructed pulmonary vascular tree data. The present invention achieves comprehensive processing of blood vessels of varying thicknesses through the innovative combination of multi-scale analysis and watershed segmentation. Multi-scale analysis can capture vascular features at different scales, while watershed segmentation can accurately locate vascular boundaries, enabling the method to simultaneously process large blood vessels such as the main pulmonary artery and tiny peri-alveolar capillaries.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing technology, and more particularly to a pulmonary vascular tree reconstruction method based on multi-scale watershed segmentation. Background Art

[0002] In modern medical imaging, accurate reconstruction of the pulmonary vascular tree is crucial for the diagnosis, treatment planning, and prognosis of lung diseases. Advances in computed tomography (CT) technology have enabled precise reconstruction of the pulmonary vascular tree. However, due to the complexity and diversity of the pulmonary vascular structure, achieving high-precision reconstruction remains challenging.

[0003] Traditional pulmonary vascular tree reconstruction methods primarily include threshold segmentation, region growing, and model matching. These methods perform adequately for simple vascular structures, but often suffer from numerous shortcomings when dealing with complex pulmonary vascular networks. For example, threshold segmentation struggles with image noise and tissue heterogeneity, leading to over-segmentation or under-segmentation. Region growing is prone to leaks or interruptions when dealing with bifurcations and small vessels. Model matching suffers from low accuracy when dealing with vessels exhibiting significant morphological variation.

[0004] In recent years, with the advancement of machine learning and deep learning technologies, convolutional neural network-based vascular segmentation methods have made some progress. However, these methods often require large amounts of labeled data for training and are not stable when dealing with vessels of varying thicknesses. Furthermore, purely data-driven methods struggle to fully leverage prior knowledge of human vascular structure and, in some complex situations, may produce reconstructions that do not conform to anatomical rules.

[0005] The closest approach in the existing technology is a vessel segmentation algorithm based on multi-scale analysis. By analyzing image features at different scales, these methods can effectively handle vessels of varying thicknesses. However, these methods still have difficulties dealing with complex situations such as vascular bifurcations and overlapping vessels. Especially when dealing with small pulmonary vessels, since their diameters approach the resolution limit of CT images, segmentation discontinuities or missed detections are prone to occur. In addition, existing multi-scale methods often struggle to effectively balance computational efficiency and reconstruction accuracy, and face the challenge of insufficient real-time performance in clinical applications.

[0006] In summary, existing pulmonary vascular tree reconstruction methods still have significant shortcomings in handling complex vascular structures, balancing the reconstruction accuracy of vessels at different scales, and ensuring the anatomical plausibility of the reconstruction results. Therefore, new pulmonary vascular tree reconstruction methods are urgently needed that can comprehensively consider multi-scale characteristics, ensure the continuity and integrity of the reconstruction results, and simultaneously take into account computational efficiency. Summary of the Invention

[0007] This paper aims to address the aforementioned technical issues by proposing a pulmonary vascular tree reconstruction method based on multi-scale watershed segmentation. This method innovatively combines multi-scale analysis with a watershed segmentation algorithm to achieve precise segmentation and reconstruction of vessels of varying thicknesses. Furthermore, by incorporating an adaptive optimization mechanism, this method effectively improves computational efficiency while maintaining reconstruction accuracy, meeting the demands of clinical applications.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0009] Pulmonary vascular tree reconstruction method based on multi-scale watershed segmentation,

[0010] The acquisition steps include:

[0011] Acquire lung CT image data;

[0012] Processing steps include:

[0013] performing an image preprocessing operation based on the lung CT image data to obtain preprocessed image data;

[0014] Performing multi-scale analysis based on the pre-processed image data to obtain multi-scale feature data;

[0015] Based on the multi-scale feature data, executing a watershed segmentation algorithm to obtain a blood vessel segmentation result;

[0016] Based on the blood vessel segmentation result, performing a blood vessel tree reconstruction operation to obtain reconstructed pulmonary blood vessel tree data;

[0017] Output steps include:

[0018] The reconstructed pulmonary vascular tree data is output.

[0019] Preferably, the image preprocessing operation specifically includes:

[0020] Based on the lung CT image data, performing a 3DSobel edge extraction operation to obtain edge-enhanced image data;

[0021] Based on the edge-enhanced image data, performing gradient amplitude calculation to obtain a gradient image;

[0022] Based on the gradient image, adaptive threshold processing is performed to obtain a binary image.

[0023] Preferably, the multi-scale analysis specifically includes:

[0024] Calculate the thinness ratio and roundness ratio of blood vessel shape;

[0025] Construct scale space based on Gaussian kernel function;

[0026] In the scale space, multi-scale feature extraction is performed.

[0027] Preferably, the watershed segmentation algorithm specifically includes:

[0028] Divide the initial blood vessel edge into multiple regions;

[0029] Construct the evolution function of the gradient in scale space;

[0030] Calculating the integral of the evolution function over a specified interval;

[0031] Based on the integration result, a watershed point is determined.

[0032] Preferably, the specified interval is [0,1];

[0033] The evolution function is p(u);

[0034] The integration result is I(u).

[0035] Preferably, the vascular tree reconstruction operation specifically includes:

[0036] Perform morphological processing to remove noise and small areas;

[0037] Extract the blood vessel centerline;

[0038] Constructing the vascular tree topology;

[0039] The blood vessel diameter information is restored through the interpolation algorithm.

[0040] As an advantage, a multi-scale fusion step is also included:

[0041] Perform segmentation and reconstruction operations at different scales to obtain reconstruction results at multiple scales;

[0042] A weighted fusion strategy is adopted to merge the reconstruction results of the multiple scales to obtain fused vascular tree data.

[0043] As an advantage, it also includes the evaluation step:

[0044] evaluating the reconstructed pulmonary vascular tree data based on preset evaluation indicators;

[0045] The evaluation indicators include segmentation accuracy and reconstruction completeness.

[0046] As an optimization step, it also includes:

[0047] Adjusting algorithm parameters based on the evaluation results;

[0048] The processing steps are re-executed using the adjusted algorithm parameters.

[0049] Preferably, the reconstructed pulmonary vascular tree data includes:

[0050] Vascular spatial location information;

[0051] Blood vessel diameter information;

[0052] Vascular branching structure information.

[0053] The method of the present invention has the following significant technical effects:

[0054] First, this method achieves comprehensive processing of vessels of varying sizes through the innovative combination of multi-scale analysis and watershed segmentation. Multi-scale analysis captures vascular characteristics at varying scales, while watershed segmentation precisely locates vessel boundaries. The synergistic effect of these two techniques enables this method to simultaneously process large vessels, such as the main pulmonary artery, and tiny peri-alveolar capillaries, significantly improving the completeness and accuracy of the reconstruction results.

[0055] Secondly, the adaptive thresholding and gradient evolution function introduced in this invention effectively address the difficulties faced by traditional methods in processing complex vascular structures. In particular, in areas such as bifurcations and overlaps, this method accurately identifies vascular boundaries, avoiding over-segmentation and under-segmentation. This characteristic is crucial for accurately reconstructing complex interlobular vascular networks.

[0056] Furthermore, the method of the present invention effectively ensures the continuity and anatomical plausibility of the reconstruction results by incorporating steps such as morphological processing and vessel centerline extraction. This not only improves the reliability of the reconstruction results but also provides a more accurate data foundation for subsequent hemodynamic analysis. For example, accurate information on vessel diameter and branching structure is crucial in the assessment of pulmonary hypertension.

[0057] Furthermore, the multi-scale fusion strategy and adaptive optimization mechanism of the present invention achieve a good balance between reconstruction accuracy and computational efficiency. This enables the method to meet the needs of real-time clinical diagnosis and provide timely imaging support for diseases requiring rapid diagnosis, such as acute pulmonary embolism.

[0058] Finally, the comprehensive vascular tree data output by this method provides rich information for subsequent medical research and clinical applications. For example, precise vascular spatial location and diameter information can be used in lung cancer surgical planning, helping doctors determine the optimal resection strategy. Detailed branching information can be used to assess pulmonary vascular malformations and provide guidance for interventional treatment.

[0059] In summary, the multi-scale watershed segmentation-based pulmonary vascular tree reconstruction method proposed in this paper effectively addresses existing challenges through the integration of multiple innovative technologies, significantly improving the accuracy, completeness, and efficiency of pulmonary vascular tree reconstruction. This not only provides a powerful tool for related medical research but also significantly contributes to improving the diagnosis and treatment of lung diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 The figure is an overall flow chart of the method of the present invention.

[0061] Figure 2 This is a flowchart of the image preprocessing of the present invention.

[0062] Figure 3 Flowchart of the multi-scale analysis of the present invention.

[0063] Figure 4 This is a flow chart of the watershed segmentation of the present invention.

[0064] Figure 5 Flowchart of the vessel tree reconstruction of the present invention.

[0065] Figure 6 Flowchart of the evaluation and optimization process of the present invention. DETAILED DESCRIPTION

[0066] like Figure 1-6 As shown, the present invention provides a method for reconstructing the pulmonary vascular tree based on multi-scale watershed segmentation. This method achieves accurate extraction and reconstruction of vascular structures in lung CT images by innovatively combining multi-scale analysis and watershed segmentation algorithms. The technical solutions of the present invention are described in detail below.

[0067] First, the method of the present invention includes an acquisition step, a processing step, and an output step. In the acquisition step, the method acquires lung CT image data. This CT image data is typically acquired by CT scanning equipment in medical institutions and contains detailed structural information about the patient's lungs.

[0068] Next, in the processing step, the method first performs an image preprocessing operation based on the acquired lung CT image data to obtain preprocessed image data. The image preprocessing operation is a very critical step, which lays the foundation for subsequent analysis and processing. Preferably, in one embodiment of the present invention, the image preprocessing operation specifically includes the following steps:

[0069] First, a 3D Sobel edge extraction operation is performed on the lung CT image data to obtain edge-enhanced image data. The 3D Sobel operator is an extension of the traditional 2D Sobel operator in three-dimensional space, which can better capture the three-dimensional structural information in CT images. The mathematical expression of the 3D Sobel operator is as follows:

[0070] ,

[0071] ,

[0072] ,

[0073] in, is the input 3D image data, Represents a convolution operation.

[0074] Then, based on the edge-enhanced image data, a gradient magnitude calculation is performed to obtain a gradient image. The calculation formula of the gradient magnitude is:

[0075] ,

[0076] Finally, based on the gradient image, adaptive thresholding is performed to obtain a binary image. The selection of the adaptive threshold is usually based on the histogram distribution of the image. A commonly used method is the Otsu method, which determines the optimal threshold by maximizing the between-class variance.

[0077] After preprocessing is completed, the method performs multi-scale analysis based on the preprocessed image data to obtain multi-scale feature data. Multi-scale analysis is an innovation of the present invention, which can simultaneously process vascular structures of different thicknesses. In a preferred embodiment of the present invention, the multi-scale analysis specifically includes the following steps:

[0078] First, calculate the thinness ratio and roundness ratio of the blood vessel shape. Thinness ratio and roundness ratio The calculation formula is as follows:

[0079] ,

[0080] ,

[0081] Where L is the length of the blood vessel, W is the width of the blood vessel, A is the cross-sectional area of the blood vessel, and P is the circumference of the blood vessel.

[0082] Next, we construct the scale space based on the Gaussian kernel function. The expression of the Gaussian kernel function is:

[0083] ,

[0084] in, is the scale parameter that controls the width of the Gaussian kernel.

[0085] Finally, multi-scale feature extraction is performed in the scale space. This step effectively captures vascular structures of different sizes by analyzing image features at different scales.

[0086] After the multi-scale analysis is completed, this method executes a watershed segmentation algorithm based on the multi-scale feature data to obtain a blood vessel segmentation result. The watershed segmentation algorithm is an image segmentation method based on topological theory. It treats the image as a topographic map and achieves segmentation by simulating water flow. In the present invention, the specific implementation steps of the watershed segmentation algorithm are as follows:

[0087] First, the initial blood vessel edge is divided into multiple regions. This step is achieved by detecting local minimum points.

[0088] Then, the evolution function of the gradient in the scale space is constructed. It describes the law of gradient changing with scale, and its general form can be expressed as:

[0089] ,

[0090] in, is the image gradient, is the scale parameter.

[0091] Next, the integral of the evolution function over the specified interval is calculated. In the present invention, the specified interval is usually selected as [0,1]. The calculation formula is:

[0092] ,

[0093] Finally, based on the integration results, the watershed point is determined. The determination of the watershed point is usually based on local maximum of .

[0094] Through the above steps, this method can accurately segment the vascular structures in lung CT images. This multi-scale watershed segmentation method has higher accuracy and robustness than traditional methods, especially when dealing with complex vascular branching structures.

[0095] Finally, in the output step, the method outputs the reconstructed pulmonary vascular tree data. This data contains information such as the spatial location, diameter, and branching structure of the vessels, providing an important basis for subsequent medical diagnosis and analysis.

[0096] The detailed description above demonstrates the significant technical advantages and innovations of the present method in the field of pulmonary vascular tree reconstruction. It not only improves the accuracy of vessel segmentation and reconstruction but also effectively processes vascular structures of varying scales, providing strong support for related medical research and clinical applications. Building on the aforementioned technical solutions, the present method also includes several key steps and technical features to further enhance the accuracy and effectiveness of pulmonary vascular tree reconstruction.

[0097] First, the present invention provides a more detailed implementation of the watershed segmentation algorithm. In a preferred embodiment of the present invention, the watershed segmentation algorithm specifically includes the following steps:

[0098] First, the initial vascular edge is divided into multiple regions. This step aims to decompose the complex vascular network into relatively simple subregions, facilitating subsequent refinement. The segmentation method is typically based on local image features, such as grayscale values and gradients. Preferably, erosion or dilation operations, which are morphological operations, can be used to assist in region segmentation.

[0099] Next, the evolution function of the gradient in the scale space is constructed. This step is the core of the watershed segmentation algorithm, which describes the law of image gradient changes with scale. In the embodiment of the present invention, the evolution function The specific form can be expressed as:

[0100] ,

[0101] in, represents the image gradient, The scale is Gaussian kernel function, Represents the convolution operation. This evolution function can effectively capture the gradient change characteristics of the image at different scales.

[0102] Then, the integral of the evolution function on the specified interval is calculated. In the present invention, the specified interval is selected as [0, 1]. This choice is based on a large number of experiments and experiences, and it can achieve a good balance between computational efficiency and accuracy. The calculation formula is:

[0103] ,

[0104] This integral result reflects the cumulative effect of the gradient changes of the image in the entire scale space.

[0105] Finally, based on the integration result, the watershed point is determined. The determination of the watershed point is a key step in the watershed segmentation algorithm, which directly affects the final segmentation effect. In the method of the present invention, the determination of the watershed point is based on Specifically, if a point The value is greater than all points in its neighborhood value, the point is marked as a watershed point.

[0106] In order to further improve the robustness of the algorithm, the present invention also introduces some parameters to control the selection of watershed points. For example, a threshold T can be set, and only when When the local maximum of exceeds T, the point is marked as a watershed point. The choice of this threshold is usually determined by the characteristics of the image and the specific application scenario. In the application of pulmonary vascular tree reconstruction, based on a large amount of experimental data, it is found that setting T to Around 70% of the maximum value can achieve good results.

[0107] Through the above steps, the watershed segmentation algorithm of the present invention can effectively segment the vascular structure in lung CT images. Compared with traditional single-scale methods, this multi-scale watershed segmentation method can better handle blood vessels of different thicknesses, thereby improving the accuracy and completeness of segmentation.

[0108] After the watershed segmentation is completed, the method of the present invention further includes a vessel tree reconstruction operation. In one embodiment of the present invention, the vessel tree reconstruction operation specifically includes the following steps:

[0109] First, morphological processing is performed to remove noise and small regions. This step aims to improve the quality and reliability of the reconstruction results. Common morphological operations include opening and closing, which can effectively remove small noisy regions while maintaining the integrity of the main vascular structures.

[0110] Next, the vessel centerlines are extracted. Extracting the vessel centerlines is a key step in reconstructing the vascular tree, providing the framework for subsequent topological structure construction. This method utilizes a distance transform-based centerline extraction algorithm that accurately captures the center position of the vessel, even in complex areas with bends or bifurcations.

[0111] Next, the vascular tree topology is constructed. This step builds a complete vascular tree structure based on the extracted centerlines by analyzing the connectivity and branching relationships of the vessels. During this construction process, the present invention employs an adaptive connection strategy that effectively addresses breakage issues caused by image noise or segmentation errors.

[0112] Finally, an interpolation algorithm is used to restore the vessel diameter information. This vessel diameter information is crucial for subsequent hemodynamic analysis. The present invention uses an interpolation algorithm based on radial basis functions, which can smoothly restore changes in vessel diameter and avoid the step effect that can be introduced by traditional linear interpolation methods.

[0113] In order to further improve the accuracy and completeness of the reconstruction results, the present invention also introduces a multi-scale fusion step. In a preferred embodiment of the present invention, the multi-scale fusion step specifically includes:

[0114] First, segmentation and reconstruction are performed at different scales to obtain reconstruction results at multiple scales. The different scales here typically refer to different spatial resolutions or different characteristic scales. For example, three to five different scales can be selected, corresponding to the characteristic scales of large, medium, and small vessels, respectively.

[0115] Then, a weighted fusion strategy is used to merge the reconstruction results of the multiple scales to obtain the fused vascular tree data. The core of the weighted fusion strategy is to assign appropriate weights to the results of each scale. In one embodiment of the present invention, the weight assignment is based on the reliability and completeness of the results of each scale. Specifically, the following weighting formula can be used:

[0116] ,

[0117] in, is the result of fusion, is the reconstruction result of the i-th scale, is the corresponding weight, and n is the number of scales. This can be determined by evaluating the quality of the results at each scale, for example, it can be calculated based on indicators such as vessel continuity and branch integrity.

[0118] Through this multi-scale fusion method, the present invention can effectively combine the reconstruction advantages at different scales, ensuring the overall structure of large blood vessels while accurately capturing the details of small blood vessels, thereby obtaining a more complete and accurate pulmonary vascular tree reconstruction result.

[0119] The above describes in detail the specific implementations of the present invention in terms of the watershed segmentation algorithm, vascular tree reconstruction, and multi-scale fusion. The organic combination of these technical features gives the present method significant advantages in the field of pulmonary vascular tree reconstruction, providing more reliable and accurate data support for related medical research and clinical applications.

[0120] Building on the aforementioned technical solutions, the present invention also introduces evaluation and optimization mechanisms to further improve the quality and reliability of pulmonary vascular tree reconstruction. These mechanisms not only objectively evaluate the reconstruction results but also adaptively adjust algorithm parameters based on the evaluation results, thereby achieving continuous optimization of the method.

[0121] First, the method of the present invention includes an evaluation step. In a preferred embodiment of the present invention, the evaluation step specifically includes evaluating the reconstructed pulmonary vascular tree data based on preset evaluation metrics. These evaluation metrics include segmentation accuracy and reconstruction completeness. These evaluation metrics are selected based on the specific requirements of pulmonary vascular tree reconstruction and clinical application needs.

[0122] The segmentation accuracy is usually calculated by comparing with the gold standard of manual annotation. In an embodiment of the present invention, the Dice coefficient can be used to quantify the accuracy of segmentation. The calculation formula of the Dice coefficient is as follows:

[0123] ,

[0124] Here, X represents the algorithm segmentation result, and Y represents the gold standard of manual annotation. The Dice coefficient ranges from 0 to 1, with larger values indicating more accurate segmentation results. Based on extensive experimental data, a Dice coefficient of 0.85 or higher is generally considered a good result for pulmonary vascular tree segmentation.

[0125] Reconstruction integrity focuses on the integrity and continuity of the vascular tree structure. In one embodiment of the present invention, reconstruction integrity can be evaluated through the following aspects:

[0126] 1. Number of vascular branches: Compared with anatomical knowledge, the reconstruction results are evaluated to see whether they capture enough vascular branches.

[0127] 2. Vascular continuity: Detects whether there are unreasonable breaks or gaps in the reconstruction results.

[0128] 3. Vascular diameter consistency: Evaluate whether the changes in vascular diameter conform to physiological laws.

[0129] Preferably, the present invention also introduces a comprehensive scoring mechanism that combines the above indicators to obtain an overall score. This comprehensive score can more comprehensively reflect the quality of the reconstruction results. For example, the following weighted summation formula can be used:

[0130] ,

[0131] in, are the weights of various indicators, which can be adjusted according to the specific application scenario. In one embodiment of the present invention, these weights can be set as: =0.4, =0.2, =0.2, =0.2. This weight distribution takes into account both the importance of segmentation accuracy and the influence of other structural features.

[0132] After the evaluation step, the method of the present invention further comprises an optimization step. In one embodiment of the present invention, the optimization step specifically comprises: adjusting algorithm parameters based on the evaluation results; and re-performing the processing step using the adjusted algorithm parameters.

[0133] This optimization step actually constitutes a closed-loop feedback mechanism that can continuously improve the quality of the reconstruction results. Specifically, the following key parameters can be automatically adjusted based on the evaluation results:

[0134] 1. Edge enhancement intensity in preprocessing: If the detection effect of small blood vessels is found to be poor, the intensity of edge enhancement can be appropriately increased.

[0135] 2. Scale range in multi-scale analysis: Adjust the scale range of multi-scale analysis based on the integrity score of vascular branches.

[0136] 3. Threshold parameters in watershed segmentation: Fine-tune the threshold parameters in the watershed segmentation algorithm based on the segmentation accuracy.

[0137] 4. Connection strategy parameters in vascular tree reconstruction: Adjust the connection strategy parameters based on the vascular continuity score.

[0138] Preferably, the present invention uses a parameter optimization algorithm based on gradient descent. The algorithm adjusts the parameters in an iterative manner so that the evaluation score is continuously improved. The iterative process can be expressed as:

[0139] ,

[0140] in, Represents the parameter set of the tth iteration is the learning rate, is the negative of the evaluation score, represents the gradient of the evaluation score with respect to the parameters.

[0141] In this way, the method of the present invention can adaptively optimize algorithm performance and continuously improve the quality of reconstruction results. This self-optimization mechanism makes the method highly adaptable and can process various CT images generated by different patients and different scanning equipment.

[0142] Finally, the present invention clearly defines the content of the reconstructed pulmonary vascular tree data. In one embodiment of the present invention, the reconstructed pulmonary vascular tree data includes: vascular spatial position information, vascular diameter information, and vascular branch structure information.

[0143] The spatial location of blood vessels is typically represented as three-dimensional coordinates, accurately describing their distribution within the lung space. Preferably, these coordinates can be based on the patient coordinate system defined in the DICOM standard, facilitating registration and fusion with other medical imaging data.

[0144] Vessel diameter information reflects changes in vessel thickness. In embodiments of the present invention, vessel diameter is typically recorded at regular intervals (e.g., 1 mm) along the vessel centerline. This representation method accurately reflects vessel morphology while maintaining data compactness.

[0145] The vascular branching structure information describes the topology of the vascular tree. Preferably, it can be represented using a tree data structure, where each node contains information such as the location of the vascular branch point and the number of the connected sub-branch. This representation method facilitates subsequent hemodynamic analysis and lesion localization.

[0146] By providing this rich information, the pulmonary vascular tree data reconstructed by the present invention provides a solid foundation for various medical applications. For example, this data can be used for automatic detection of pulmonary embolism, assessment of pulmonary hypertension, and surgical planning.

[0147] In summary, this invention further improves the quality and practicality of pulmonary vascular tree reconstruction by introducing evaluation and optimization mechanisms and clearly defining the content of reconstruction data. The combination of these technical features gives the method significant advantages in accuracy, reliability, and adaptability, providing strong support for related medical research and clinical applications.

[0148] The above description is only a preferred specific embodiment of the present invention; however, the protection scope of the present invention is not limited thereto; any person familiar with the art who makes equivalent replacements or changes based on the scheme and improved concepts of the present invention within the scope disclosed by the present invention shall be covered by the protection scope of the present invention.

Claims

1. A pulmonary vascular tree reconstruction method based on multi-scale watershed segmentation, characterized by: The acquisition steps include: Acquire lung CT image data; Processing steps include: performing an image preprocessing operation based on the lung CT image data to obtain preprocessed image data; Performing multi-scale analysis based on the pre-processed image data to obtain multi-scale feature data; Based on the multi-scale feature data, executing a watershed segmentation algorithm to obtain a blood vessel segmentation result; Based on the blood vessel segmentation result, performing a blood vessel tree reconstruction operation to obtain reconstructed pulmonary blood vessel tree data; Output steps include: outputting the reconstructed pulmonary vascular tree data; The watershed segmentation algorithm specifically includes: Divide the initial blood vessel edge into multiple regions; Construct the evolution function of the gradient in scale space; Calculating the integral of the evolution function over a specified interval; Based on the integration result, a watershed point is determined.

2. The method according to claim 1, characterized in that The image preprocessing operation specifically includes: Based on the lung CT image data, performing a 3DSobel edge extraction operation to obtain edge-enhanced image data; Based on the edge-enhanced image data, performing gradient amplitude calculation to obtain a gradient image; Based on the gradient image, adaptive threshold processing is performed to obtain a binary image.

3. The method according to claim 1, characterized in that The multi-scale analysis specifically includes: Calculate the thinness ratio and roundness ratio of blood vessel shape; Construct scale space based on Gaussian kernel function; In the scale space, multi-scale feature extraction is performed.

4. The method according to claim 1, wherein: The specified interval is [0,1]; The evolution function is p(u); The integration result is I(u), where u is a scale parameter variable used to represent changes in the scale space.

5. The method according to claim 1, wherein The vascular tree reconstruction operation specifically includes: Perform morphological processing to remove noise and small areas; Extract the blood vessel centerline; Constructing the vascular tree topology; The blood vessel diameter information is restored through the interpolation algorithm.

6. The method according to claim 1, wherein It also includes a multi-scale fusion step: Perform segmentation and reconstruction operations at different scales to obtain reconstruction results at multiple scales; A weighted fusion strategy is adopted to merge the reconstruction results of the multiple scales to obtain fused vascular tree data.

7. The method according to claim 1, characterized in that Also includes the evaluation step: evaluating the reconstructed pulmonary vascular tree data based on preset evaluation indicators; The evaluation indicators include segmentation accuracy and reconstruction completeness.

8. The method according to claim 7, characterized in that Also includes optimization steps: Adjusting algorithm parameters based on the evaluation results; The processing steps are re-executed using the adjusted algorithm parameters.

9. The method according to claim 1, characterized in that The reconstructed pulmonary vascular tree data includes: Vascular spatial location information; Blood vessel diameter information; Vascular branching structure information.

Images