Method and system for analyzing body composition based on abdomen CT (Computed Tomography) scanning

Abstract

The invention discloses a method and a system for analyzing body components based on abdominal CT (computed tomography) scanning, which effectively solve the problem of spine confusion caused by highly similar structures of adjacent vertebral bodies by introducing a local anatomical marker on an image, and do not need to depend on a large data set to perform remote vertebral body context learning, thereby reducing the data demand and improving the accuracy of the data. And the reliability of spine segmentation is improved. According to the method, a two-stage segmentation strategy is adopted, spine segmentation and body composition segmentation are optimized respectively, the independence and precision of each segmentation task are guaranteed, the strategy can effectively improve the segmentation accuracy, and compared with an existing method, centrum confusion is reduced, and the consistency of measurement results is improved. According to the method, three-dimensional volume measurement of muscle, subcutaneous fat and visceral fat is realized, and compared with a single-slice or multi-slice two-dimensional area inference method, tissue distribution in a whole spine range can be reflected more comprehensively and accurately, so that the volume measurement error is remarkably reduced, and the reference of data is improved.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and in particular to a method and system for analyzing body composition based on abdominal CT scans. Background Technology

[0002] With the development of medical imaging technology, CT has become a widely used routine examination method in clinical practice. Body composition assessment based on CT images requires no additional image acquisition and can simultaneously obtain quantitative information on tissues such as skeletal muscle, subcutaneous adipose tissue (SAT), and visceral adipose tissue (VAT). This helps to assess a patient's nutritional status, disease progression, surgical risks, and long-term prognosis, and therefore has important clinical value.

[0003] In existing technologies, methods for analyzing body composition based on CT scans can be mainly divided into three categories.

[0004] The first category is body composition assessment techniques based on single-slice sections. This method typically selects a single cross-section at a specific vertebral level, such as the third lumbar vertebra (L3), to measure the area of ​​muscle and adipose tissue. The process includes slice selection, manual or automated tissue segmentation, and calculation of the area of ​​each tissue. This method is simple to operate, computationally inexpensive, and convenient for clinical application. However, its limitation lies in its inability to comprehensively reflect the distribution of muscle and fat throughout the body; single-slice representativeness is limited, and the accuracy of volume assessment is low. Inaccurate volume assessment will further lead to a series of application problems, such as biasing subsequent data users' judgments of body composition status. Furthermore, due to insufficient precision, this method is limited in its applicability in long-term follow-up studies, multi-center clinical trials, and research environments that rely on standardized body composition indicators.

[0005] The second category is multi-slice or multi-vertebral level area measurement methods. This method selects multiple slices at the vertebral level to segment and measure the area of ​​the tissue, improving the representativeness of the analysis. The process generally includes multi-slice localization, tissue segmentation, area calculation, and data aggregation. Compared to single-slice methods, multi-slice methods have a wider coverage and can reduce the bias introduced by single slices. However, measurement accuracy can still be affected by differences between slices, and some slice selections may miss important tissue distribution changes. Furthermore, it is still difficult to achieve a complete three-dimensional volume assessment. In practical use, this method also faces certain implementation limitations. On the one hand, different institutions or operators may still have differences in slice selection strategies, leading to lower consistency and reproducibility of measurement results, thus affecting cross-center studies and clinical data standardization. On the other hand, the multi-slice segmentation and index aggregation process is time-consuming and relies on more manual or semi-automated steps, increasing process complexity and computational costs, which is not conducive to rapid clinical analysis or large-scale application.

[0006] The third category is deep learning-based automatic spine segmentation techniques. These methods use convolutional neural networks or Transformer models to automatically locate and segment the spine, generating multi-label or instance segmentation results. A typical workflow includes spine localization, feature extraction, semantic segmentation or instance segmentation, and result merging. This method is highly automated, reducing manual intervention and improving the accuracy of some vertebral segmentation. However, its limitations include the fact that localization and segmentation are usually processed independently, ignoring the correlation between them; due to the high similarity of the anatomical structures of adjacent vertebrae, voxel-level misclassification, i.e., spine confusion, is prone to occur; furthermore, the lack of explicit reference information or landmarks limits the model's performance and interpretability. In summary, although this method improves automation and the accuracy of some vertebral segmentation, it still suffers from problems such as inconsistency between localization and segmentation, lack of reference information, poor cross-device adaptability, and high dependence on computational resources in practical applications. These factors limit its usability and reliability in clinical, research, and large-scale multi-center applications.

[0007] Overall, existing technologies still have significant shortcomings in body composition assessment and vertebral segmentation. Single-slice methods are limited in representativeness, have low volume assessment accuracy, and can bias subsequent judgments of body composition status by data users. While multi-slice methods have a wider coverage, their measurement accuracy is affected by differences in slice selection, and the process is complex and time-consuming, making it difficult to achieve complete three-dimensional volume assessment. Deep learning-based automatic segmentation methods have a high degree of automation, but localization and segmentation may be inconsistent, adjacent vertebrae are easily confused, explicit reference information is lacking, and they are highly dependent on computational resources, with limited cross-device adaptability and interpretability. These limitations restrict the usability and reliability of existing technologies in clinical applications, scientific research analysis, and large-scale multicenter studies.

[0008] Therefore, it is necessary to develop a method or system for analyzing body composition based on abdominal CT scans to solve the above problems. Summary of the Invention

[0009] The purpose of this invention is to design a method and system for analyzing body composition based on abdominal CT scans in order to solve the above-mentioned problems.

[0010] The present invention achieves the above objectives through the following technical solutions:

[0011] A method for analyzing body composition based on abdominal CT scans, comprising the following steps:

[0012] S1. Image preprocessing;

[0013] S2. Preliminary spinal segmentation: The preprocessed image is input into a convolutional neural network or a Transformer-based segmentation model to generate preliminary segmentation results for the T11–L5 vertebrae; by calculating the segmentation of each vertebra... Coefficients and information entropy quantify the degree of confusion for each cone;

[0014] S3. Selection and guidance of anatomical landmarks; Based on the degree of confusion score, select the first part of the vertebral body with the lowest degree of confusion as the local anatomical reference landmark; Mark the local anatomical reference landmark in the original CT image as the input for the second step of segmentation;

[0015] S4. Marker-guided fine segmentation: On the original CT image with added local anatomical reference marks, semantic segmentation of the remaining vertebrae excluding the first part is performed again to obtain the second-step fine segmentation result; the preliminary segmentation result and the second-step fine segmentation result are merged. If the voxel values ​​conflict, the preliminary segmentation result shall prevail.

[0016] S5. Post-processing of spinal segmentation yields the final spinal segmentation mask, which serves as the input for body component segmentation.

[0017] S6, Body composition segmentation.

[0018] Specifically, S1 includes: resampling the original CT image to a uniform size; limiting the CT grayscale value range to between 200 and 1000, and retaining only bone tissue features; the formula is as follows: ;

[0019] in This is the lower limit of the grayscale value, set to 200; This is the upper limit of the grayscale value, set to 1000, where h is the initial grayscale value. This is the final grayscale value.

[0020] Specifically, in S2, by calculating each vertebral body Coefficients and information entropy, quantifying the degree of confusion for each cone, specifically include three categories:

[0021] Oversegmentation: The vertebral segmentation exceeds the target range;

[0022] Incomplete segmentation: The vertebral body was not segmented;

[0023] Mild confusion: The overall number of vertebrae is correct, but a small number of voxels are misclassified as adjacent vertebrae;

[0024] The i-th vertebra The formula for calculating the coefficient is as follows:

[0025] ;

[0026] in : The set of actual labeled voxels for the i-th vertebra; The predicted binary mask for the i-th vertebra can also be obtained from the probability map by applying a threshold. It is a very small constant;

[0027] Information entropy The calculation formula is as follows:

[0028] ;

[0029] in, Represents the total number of voxels. voxels The predicted probability at that location;

[0030] level of confusion The definition is as follows:

[0031] .

[0032] Specifically, in S5, post-processing of the spine segmentation includes:

[0033] Segmented processing along the sagittal axis: retain the region containing the most target vertebral voxels and exclude non-target regions;

[0034] Cross-sectional section processing: If multiple vertebral body labels exist in a section, they are uniformly assigned to the label that appears most frequently;

[0035] Segmented processing along the vertical axis: Similar to the sagittal axis processing method, adjust the label values ​​of other regions to ensure the independence and integrity of each vertebra.

[0036] Specifically, in S6, body composition segmentation includes:

[0037] Input image preparation;

[0038] The original CT image is stitched together with the spinal segmentation mask obtained from S5 according to channel to form a multi-channel input, providing anatomical reference information. The formula is as follows:

[0039]

[0040] in As input to the body composition segmentation model, As a spinal segmentation mask obtained from S5;

[0041] Volume segmentation;

[0042] Three-dimensional volume segmentation of skeletal muscle, subcutaneous fat and visceral fat was performed to repair the loss of spinal boundary details caused by the first stage of preprocessing;

[0043] Volume calculation and result output;

[0044] The three-dimensional volume segmentation results are voxel counted and multiplied by the voxel volume to obtain the three-dimensional volume of each tissue; the volume of skeletal muscle, subcutaneous fat and visceral fat within each vertebral body is output.

[0045] A system for analyzing body composition based on abdominal CT scans, comprising:

[0046] Image preprocessing module;

[0047] The preliminary spine segmentation module is used to input the preprocessed image into a convolutional neural network or a Transformer-based segmentation model to generate preliminary segmentation results for the T11–L5 vertebrae; it calculates the segmentation results for each vertebra. Coefficients and information entropy quantify the degree of confusion for each cone;

[0048] Anatomical landmark selection and guidance module; used for anatomical landmark selection and guidance; based on the degree of confusion score, the first part of the vertebral body with the lowest degree of confusion is selected as the local anatomical reference landmark; the local anatomical reference landmark is marked in the original CT image as input for the second step of segmentation;

[0049] The marker-guided fine segmentation module is used to perform semantic segmentation of the remaining vertebrae after removing the first part on the original CT image with local anatomical reference markers added, to obtain the second-step fine segmentation result; the preliminary segmentation result and the second-step fine segmentation result are merged, and if the voxel values ​​conflict, the preliminary segmentation result shall prevail;

[0050] The spinal segmentation post-processing module is used for spinal segmentation post-processing to obtain the final spinal segmentation mask, which serves as the input for body component segmentation.

[0051] Body composition segmentation module.

[0052] The beneficial effects of this invention are:

[0053] This invention utilizes abdominal CT scans to achieve three-dimensional volumetric analysis of muscles, subcutaneous fat, and visceral fat within the T11 to L5 vertebral body region. By introducing local anatomical landmarks onto the images, it effectively solves the problem of spinal confusion caused by the high similarity of adjacent vertebral structures. It eliminates the need for large datasets for long-range vertebral context learning, thereby reducing data requirements and improving the reliability of spinal segmentation.

[0054] This invention employs a two-stage segmentation strategy to optimize spine segmentation and body component segmentation separately, ensuring the independence and accuracy of each segmentation task. This strategy can effectively improve segmentation accuracy, reduce vertebral body confusion compared to existing methods, and improve the consistency of measurement results.

[0055] This invention enables three-dimensional volume measurement of muscle, subcutaneous fat, and visceral fat. Compared with single-slice or multi-slice two-dimensional area inference methods, it can more comprehensively and accurately reflect the tissue distribution across the entire spine, thereby significantly reducing volume measurement errors and improving the reliability of the data.

[0056] The analytical results provided by this invention can serve as a reference tool for doctors or researchers to assess body composition, improving the consistency and reliability of the assessment, reducing data bias caused by inaccurate slice selection or segmentation, and providing reliable data support for clinical research, health management, and multi-center data analysis.

[0057] This invention has a high level of automation, reduces manual operation, and improves processing efficiency. It can be widely used in clinical research, health management and related fields, providing repeatable and accurate technical support for related research, while overcoming the limitations of existing technologies in terms of representativeness, measurement accuracy and spinal segmentation accuracy. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0060] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0061] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0062] In the description of this invention, it should be understood that the terms "upper," "lower," "inner," "outer," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used to facilitate the description of this invention and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0063] Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0064] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, terms such as "set" and "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0065] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0066] like Figure 1 As shown, a method for analyzing body composition based on abdominal CT scans is presented. This method enables three-dimensional volume measurement of skeletal muscle (SM), subcutaneous adipose tissue (SAT), and visceral adipose tissue (VAT) within the T11 to L5 vertebral body range, solving the problems of easy confusion in vertebral segmentation and inaccurate volume measurement in existing technologies. This method mainly includes two stages: a vertebral segmentation stage and a body composition segmentation stage, with dedicated preprocessing and post-processing steps in each stage to ensure accuracy.

[0067] A method for analyzing body composition based on abdominal CT scans includes:

[0068] The goal of Phase 1 vertebral segmentation is to accurately segment the T11 to L5 vertebrae, resolving the confusion caused by the similarity of adjacent vertebral anatomical structures in traditional methods, and providing reliable spinal reference information for subsequent body composition segmentation, including:

[0069] S1. Image Preprocessing: Resample the original CT image to a uniform size (e.g., 192×192×256) to ensure that the subsequent segmentation model can capture global contextual information across the entire spine. Limit the CT grayscale value range (Hounsfield Unit, HU) to between 200 and 1000, preserving only bone tissue features to reduce interference from surrounding soft tissues; the formula is as follows:

[0070] ;

[0071] in This is the lower limit of the grayscale value, set to 200; This is the upper limit of the grayscale value, set to 1000, where h is the initial grayscale value. This is the final grayscale value.

[0072] S2. Preliminary spinal segmentation: The preprocessed image is input into a convolutional neural network (CNN) or a Transformer-based segmentation model to generate preliminary segmentation results for the T11–L5 vertebrae; by calculating the segmentation of each vertebra... Coefficients and information entropy quantify the confusion degree of each cone, including three categories:

[0073] Over-segmentation: Vertebrae are segmented beyond the target range, such as T10 being misclassified as T11, leading to an increase in false positives.

[0074] Under-segmentation: The vertebral body is not segmented, such as T11, which leads to an increase in false negatives.

[0075] Minor Confusion: The overall number of vertebrae is correct, but a small number of voxels are misclassified as adjacent vertebrae;

[0076] The i-th vertebra The formula for calculating the coefficient is as follows:

[0077] ;

[0078] in : The set of actual labeled voxels for the i-th vertebra; The predicted binary mask for the i-th vertebra can also be obtained from the probability map by applying a threshold. For example, a very small constant (e.g.) );

[0079] Information entropy The calculation formula is as follows:

[0080] ;

[0081] in, Represents the total number of voxels. voxels The predicted probability at that location;

[0082] level of confusion The definition is as follows:

[0083] .

[0084] S3. Anatomical landmark selection and guidance: Based on the confusion score, select the first part of the vertebral body with the lowest confusion (such as L3–L5) as the local anatomical reference landmark; mark the local anatomical reference landmark in the original CT image (mask) as the input for the second step of segmentation, enhance the local feature representation, and improve the segmentation accuracy of adjacent vertebral bodies;

[0085] S4. Marker-guided fine segmentation: On the original CT image with added local anatomical reference marks, semantic segmentation of the remaining vertebral bodies (T11–L2) after removing the first part is performed again to obtain the second-step fine segmentation result; the preliminary segmentation result and the second-step fine segmentation result are merged. If the voxel values ​​conflict, the preliminary segmentation result shall prevail.

[0086] S5. Post-processing of spinal segmentation yields the final spinal segmentation mask, which serves as the input for body component segmentation.

[0087] Post-spinal segmentation processing includes:

[0088] Segmented processing along the sagittal axis: retain the region containing the most target vertebral voxels and exclude non-target regions;

[0089] Cross-sectional section processing: If multiple vertebral body labels exist in a section, they are uniformly assigned to the label that appears most frequently;

[0090] Segmented processing along the vertical axis: Similar to the sagittal axis processing method, adjust the label values ​​of other regions to ensure the independence and integrity of each vertebra.

[0091] The goal of Phase Two, body composition segmentation, is to perform three-dimensional volumetric segmentation of skeletal muscle, subcutaneous fat, and visceral fat in abdominal CT images, based on high-precision spine segmentation results, to achieve full-range volumetric assessment. The steps are as follows:

[0092] S6, Body composition segmentation.

[0093] Body composition segmentation includes:

[0094] Input image preparation;

[0095] The original CT image is stitched together with the spinal segmentation mask obtained from S5 according to channel to form a multi-channel input, providing anatomical reference information. The formula is as follows:

[0096]

[0097] in As input to the body composition segmentation model, As a spinal segmentation mask obtained from S5;

[0098] Volume segmentation;

[0099] Three-dimensional volume segmentation was performed on skeletal muscle (SM), subcutaneous fat (SAT), and visceral fat (VAT) to repair the loss of spinal boundary details caused by the first stage of preprocessing and ensure segmentation accuracy.

[0100] Volume calculation and result output;

[0101] The three-dimensional volume segmentation results are counted by voxels and multiplied by the voxel volume to obtain the three-dimensional volume (cm³) of each tissue; the volumes of skeletal muscle (SM), subcutaneous fat (SAT), and visceral fat (VAT) within each vertebral body are output.

[0102] A system for analyzing body composition based on abdominal CT scans, comprising:

[0103] Image preprocessing module;

[0104] The preliminary spine segmentation module is used to input the preprocessed image into a convolutional neural network or a Transformer-based segmentation model to generate preliminary segmentation results for the T11–L5 vertebrae; it calculates the segmentation results for each vertebra. Coefficients and information entropy quantify the degree of confusion for each cone;

[0105] Anatomical landmark selection and guidance module; used for anatomical landmark selection and guidance; based on the degree of confusion score, the first part of the vertebral body with the lowest degree of confusion is selected as the local anatomical reference landmark; the local anatomical reference landmark is marked in the original CT image as input for the second step of segmentation;

[0106] The marker-guided fine segmentation module is used to perform semantic segmentation of the remaining vertebrae after removing the first part on the original CT image with local anatomical reference markers added, to obtain the second-step fine segmentation result; the preliminary segmentation result and the second-step fine segmentation result are merged, and if the voxel values ​​conflict, the preliminary segmentation result shall prevail;

[0107] The spinal segmentation post-processing module is used for spinal segmentation post-processing to obtain the final spinal segmentation mask, which serves as the input for body component segmentation.

[0108] Body composition segmentation module.

[0109] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method of analyzing body composition based on abdominal CT scans, characterized by, Including the following steps: S1. Image preprocessing; S2, preliminary spinal segmentation; input the preprocessed image into a convolutional neural network or a segmentation model based on Transformer to generate a preliminary segmentation result of T11-L5 vertebral bodies; quantify the confusion degree of each vertebral body by calculating the coefficient and information entropy of each vertebral body ​ S3. Selection and guidance of anatomical landmarks: Based on the degree of confusion score, select the first part of the vertebral body with the lowest degree of confusion as the local anatomical reference landmark; Local anatomical reference landmarks are marked in the original CT image and used as input for the second step of segmentation; S4. Marker-guided fine segmentation: On the original CT image with added local anatomical reference marks, semantic segmentation of the remaining vertebrae excluding the first part is performed again to obtain the second-step fine segmentation result; the preliminary segmentation result and the second-step fine segmentation result are merged. If the voxel values ​​conflict, the preliminary segmentation result shall prevail. S5. Post-processing of spinal segmentation yields the final spinal segmentation mask, which serves as the input for body component segmentation. S6, Body composition segmentation.

2. The method of claim 1, wherein the method is based on an abdominal CT scan. S1 specifically includes: resampling the original CT image to a uniform size; limiting the CT grayscale value range to 200–1000, and retaining only bone tissue features; the formula is as follows: ； wherein is a lower limit of the limit of the gray value, set to 200; is an upper limit of the limit of the gray value, set to 1000, wherein h is an initial gray value, is a final gray value.

3. The method for analyzing body composition based on abdominal CT scan according to claim 2, characterized in that, In S2, the confusion degree of each vertebra is quantified by calculating the coefficient and information entropy of each vertebra. Specifically, it includes three categories: coefficients and information entropy of each vertebra. Specifically, it includes three categories: Oversegmentation: The vertebral segmentation exceeds the target range; Incomplete segmentation: The vertebral body was not segmented; Mild confusion: The overall number of vertebrae is correct, but a small number of voxels are misclassified as adjacent vertebrae; The i-th vertebra The formula for calculating the coefficient is as follows: ； in : The set of actual labeled voxels for the i-th vertebra; The predicted binary mask for the i-th vertebra can also be obtained from the probability map by applying a threshold. It is a very small constant; Information entropy The calculation formula is as follows: ； in, Represents the total number of voxels. voxels The predicted probability at that location; level of confusion The definition is as follows: 。 4. The method for analyzing body composition based on abdominal CT scan according to claim 3, characterized in that, In S5, post-processing of the spine segmentation includes: Segmented processing along the sagittal axis: retain the region containing the most target vertebral voxels and exclude non-target regions; Cross-sectional section processing: If multiple vertebral body labels exist in a section, they are uniformly assigned to the label that appears most frequently; Segmented processing along the vertical axis: Similar to the sagittal axis processing method, adjust the label values ​​of other regions to ensure the independence and integrity of each vertebra.

5. The method for analyzing body composition based on abdominal CT scan according to claim 4, characterized in that, In S6, body composition segmentation includes: Input image preparation; The original CT image is stitched together with the spinal segmentation mask obtained from S5 according to channel to form a multi-channel input, providing anatomical reference information. The formula is as follows: in As input to the body composition segmentation model, As a spinal segmentation mask obtained from S5; Volume segmentation; Three-dimensional volume segmentation of skeletal muscle, subcutaneous fat and visceral fat was performed to repair the loss of spinal boundary details caused by the first stage of preprocessing; Volume calculation and result output; The three-dimensional volume segmentation results are voxel counted and multiplied by the voxel volume to obtain the three-dimensional volume of each tissue; the volume of skeletal muscle, subcutaneous fat and visceral fat within each vertebral body is output.

6. A system for analyzing body composition based on abdominal CT scans, characterized in that, include: Image preprocessing module; Preliminary spinal segmentation module; This is used to input the preprocessed image into a convolutional neural network or a Transformer-based segmentation model to generate preliminary segmentation results for the T11–L5 vertebrae; by calculating each vertebra... Coefficients and information entropy quantify the degree of confusion for each cone; Anatomical landmark selection and guidance module; Used for anatomical landmark selection and guidance; based on the degree of confusion score, the first part of the vertebral body with the lowest degree of confusion is selected as the local anatomical reference landmark; Local anatomical reference landmarks are marked in the original CT image and used as input for the second step of segmentation; A finely segmented module guided by signs; This is used to re-segment the remaining vertebral bodies after removing the first part on the original CT image with local anatomical reference markers, to obtain the second-step fine segmentation result; the preliminary segmentation result and the second-step fine segmentation result are merged, and if there is a conflict in the voxel values, the preliminary segmentation result shall prevail; Spinal segmentation post-processing module; Used for post-processing of spine segmentation to obtain the final spine segmentation mask, which serves as input for body component segmentation; Body composition segmentation module.

Images