A CT image self-adaptive enhancement breast tumor rehabilitation degree evaluation system

Abstract

The present application relates to a kind of CT image adaptive enhancement breast tumor rehabilitation degree evaluation system, it is related to medical auxiliary diagnosis technical field, including image data acquisition module, adaptive enhancement module, feature perception module and rehabilitation degree quantitative evaluation module;The present application introduces the double-domain collaborative correction mechanism of projection domain adaptive interpolation and image domain GAN network, clears up radial metal artifact, overcomes the edge smooth distortion caused by traditional single projection domain repair, accurately restores the real anatomical structure of the region adjacent to implant, by mining the texture information of local imageomics, with adaptive nonlinear gray mapping algorithm, the dynamic range of soft tissue window is customized stretch, amplifies the subtle differences between postoperative fibrosis scar, edema and small recurrent tumor, while linkage historical image constructs space-time evolution model, static image observation is converted into dynamic trend tracking, provides scientific evaluation basis for clinical follow-up.

Description

Technical Field

[0001] This invention relates to the field of medical auxiliary diagnostic technology, and in particular to a breast tumor recovery assessment system based on adaptive enhancement of CT images. Background Technology

[0002] In the clinical treatment and rehabilitation monitoring of breast tumors, CT (computed tomography) is routinely used to screen for metastasis. However, existing standard CT imaging techniques have the following two problems when assessing the degree of recovery of local breast tissue: After surgical resection or radiotherapy, soft tumors are prone to local fibrosis, edema, or tissue thickening. Standard CT scans are prone to misdiagnosis or missed diagnosis when distinguishing these benign postoperative physical changes from minor local tumor recurrences because the attenuation coefficients of the two are similar and the soft tissue contrast is poor. Many breast cancer patients have silicone implants or tissue expanders implanted after surgery, or titanium clips placed at the tumor bed to assist in radiotherapy localization. High-density materials will produce severe radial metal artifacts and beam hardening effects during the reconstruction process of CT scans, covering a large area of ​​key tissues around the implants. This makes it impossible for doctors and existing computer-aided diagnostic systems to accurately assess the tissue recovery in the area adjacent to the implants and whether there are new lesions.

[0003] To address the aforementioned technical deficiencies, a solution is proposed. Summary of the Invention

[0004] The purpose of this invention is to eliminate radial metal artifacts by introducing a dual-domain collaborative correction mechanism of projection domain adaptive interpolation and image domain GAN network, thereby overcoming the edge smoothing distortion that is easily caused by traditional single projection domain repair and accurately restoring the real anatomical structure of the area adjacent to the implant.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a CT image adaptive enhancement breast tumor recovery assessment system, comprising an image data acquisition module, an adaptive enhancement module, a feature perception module, and a recovery level quantitative assessment module, wherein: The image data acquisition module is used to acquire the original projection data of the patient's postoperative breast region and the initial CT images of preliminary reconstruction, and send them to the adaptive enhancement module. The adaptive enhancement module includes a high-density material segmentation unit and a dual-domain artifact correction unit, wherein: The high-density material segmentation unit is used to acquire initial CT images, identify and extract prior information on the spatial location of the high-density material distribution area based on the initial CT images, and the high-density material includes silicone prostheses, tissue expanders and titanium clips. The dual-domain artifact correction unit utilizes the prior information of spatial location and divides the original projection domain and image domain on the initial CT image according to the original projection data. In the original projection domain, it performs adaptive interpolation repair on the projection trajectory obscured by high-density material. At the same time, in the image domain, it uses a generative adversarial network to deeply suppress residual artifacts. After obtaining the repaired projection data, it performs secondary reconstruction to generate an artifact-free basic image and sends it to the feature perception module. The feature perception module is used to receive the artifact-free base image, extract tissue texture information based on local radiomics features, use an adaptive nonlinear grayscale mapping algorithm to stretch the grayscale dynamic range within the soft tissue window, enhance the density and texture differences between fibrotic scars, edematous tissue and active tumor cell clusters, and generate a high-contrast enhanced image to be sent to the recovery degree quantification assessment module.

[0006] Furthermore, the recovery degree quantitative assessment module is used to acquire the high-contrast enhanced image to calculate the characteristic index vector of the tumor bed and marginal area after breast tumor surgery, and to construct a spatiotemporal evolution model by combining the patient's preoperative historical image sequence, and output a recovery assessment report including tissue recovery score and recurrence risk probability.

[0007] Furthermore, the specific process of delineating high-density material distribution areas is as follows: S11. Obtain the initial CT image of preliminary reconstruction, traverse the CT value of each voxel point in the initial CT image, obtain the preset material density range and perform multi-level threshold segmentation, and divide the first candidate region and the second candidate region in the initial CT image. Specifically, for silicone implants and tissue expanders, a first threshold interval is set to divide the first candidate region; For the titanium clip, a second threshold interval is set to divide the second candidate region; S12. Perform three-dimensional morphological opening and closing operations on the first candidate region and the second candidate region respectively, remove isolated bright noise points with a volume smaller than the first set threshold, and smooth the three-dimensional edge structure of the high-density candidate region.

[0008] Furthermore, the specific process for obtaining prior information about spatial location is as follows: S21. The first candidate region and the second candidate region are marked as connected components, and the geometric feature parameters of each connected component are calculated. The geometric feature parameters include the volume, surface area and shape anisotropy index of the connected component. S22. Compare the calculated geometric feature parameters with the preset chest bone atlas feature library. If the shape anisotropy index of a connected region is a slender tubular structure and the volume conforms to the human bone characteristics, then the connected region is removed from the first candidate region and the second candidate region, thereby obtaining the accurate silicone implant distribution area and the titanium clip distribution area. S23. The silicone prosthesis distribution area and the titanium clip distribution area are binarized to generate a three-dimensional spatial position mask of the high-density material. Preset CT scan geometric parameters are obtained. A forward projection operation is performed on the three-dimensional spatial position mask to map it from the three-dimensional image domain to the two-dimensional sinusoidal domain, generating a corresponding projection trajectory mask. The projection trajectory mask is the spatial position prior information, used to indicate the range of projection data that needs to be discarded and interpolated in the original projection data due to the attenuation of rays from the high-density material.

[0009] Furthermore, the specific process of adaptive interpolation repair is as follows: The original projection data is acquired, and forward projection is performed using a preset CT scan geometry matrix to map the spatial location prior information into the original projection data, thereby generating a projection trajectory mask corresponding to the rays blocked by high-density material. Using the projection trajectory mask as an index, the original projection data is divided into a reliable projection area and a damaged projection area. For data points in the damaged projection area, the data in the reliable projection area in the neighborhood of the data point is used to perform adaptive interpolation repair using a normalized metal artifact interpolation algorithm to obtain the initially repaired projection data.

[0010] Furthermore, the specific process for generating the artifact-free base image is as follows: S31. Construct a generative adversarial network model containing a generator and a discriminator. The generator adopts an encoder-decoder architecture. Its input is the intermediate transition image obtained by the first filtering back projection reconstruction of the initially repaired projection data. The generator extracts the deep features of residual water ripple artifacts and soft tissue edges through multi-layer convolution and outputs the predicted artifact distribution map. The intermediate transition image and the predicted artifact distribution map are subjected to difference operation to obtain the initial artifact-free image. S32. The discriminator is used to determine whether the texture distribution of the preliminary artifact removal image conforms to the real breast soft tissue anatomical structure. The generator network weights are continuously optimized through adversarial training until a high-fidelity image domain artifact removal feature matrix is ​​output. S33. The initially repaired projection data and the image domain artifact removal feature matrix are jointly iterated. During the secondary image reconstruction process, the image domain artifact removal feature matrix is ​​used as a regularization constraint term to perform iterative convergence reconstruction on the repaired projection data, and finally generate the artifact removal base image.

[0011] Furthermore, the specific process for generating high-contrast enhanced images is as follows: S41. Obtain the artifact-free basic image. Based on the preset breast soft tissue CT threshold window, segment the soft tissue region of interest containing the tumor bed and its edge. Calculate the three-dimensional gray-level co-occurrence matrix and local binary mode of each voxel block within the region of interest, thereby obtaining texture parameters characterizing the tissue microstructure. The texture parameters include local information entropy, second angular moment, and gray-level difference contrast. S42. Integrate the texture parameters to construct a tissue heterogeneity index. The tissue heterogeneity index is positively correlated with local information entropy and gray-level difference contrast, and negatively correlated with the second moment of angle. It is used to quantitatively distinguish between uniform edema, scar tissue and non-uniform suspected active tumor masses. S43, The tissue heterogeneity index As an adjustment parameter, an adaptive nonlinear gray-scale mapping model is constructed using an improved adaptive Sigmoid function, and the specific formula is as follows: ; in, The initial CT grayscale value of the voxel; The target grayscale value after mapping; Display the maximum value of the dynamic range for the target; This is the center offset of the soft tissue window; This is the gain coefficient; S44. Input the low-frequency background layer of the target tissue image and the artifact-removed base image into the Laplacian pyramid for multi-scale smooth fusion, and output a high-contrast enhanced image.

[0012] Furthermore, the specific process for generating a rehabilitation assessment report is as follows: S51. Acquire high-contrast enhanced images, divide the tumor bed core area and the edge extension area at a specific distance based on image threshold and morphological algorithm, extract objective physical features in the edge extension area, and calculate feature index vector. The feature index vector includes volume change index reflecting tissue proliferation / absorption, boundary sharpness index reflecting interface stability, and three-dimensional gray-level gradient variance reflecting microvascular density. S52. Obtain the preoperative and postoperative historical image sequences from the patient's electronic medical record. Based on the B-spline non-rigid registration algorithm, using the patient's stable sternum and rib contours as strong constraint reference points, elastically register and deform the historical image sequences with the current high-contrast enhanced images. Stack the feature index vectors of the registered images at each time point in chronological order to construct a multi-dimensional spatiotemporal feature sequence map X = { , , ...,}, where t is the current time node; S53. Input the spatiotemporal feature sequence map into a pre-trained spatiotemporal evolution model. The spatiotemporal evolution model uses a convolutional long short-term memory network. Through multi-layer forward propagation of the convolutional long short-term memory network, a deep spatiotemporal fusion feature matrix representing the dynamic evolution law of the organization is extracted. Deep spatiotemporal fusion feature matrix The input is fed into the terminal double-branch fully connected layer of the convolutional long short-term memory network. The terminal double-branch fully connected layer includes a first branch and a second branch: The first branch performs a regression task, calculating the current tissue recovery score using the logistic activation function. The formula is as follows: ; The second branch performs the classification probability prediction task, calculating the relapse risk probability using a normalized exponential function. The formula is as follows: ; in, , For the preset weight matrix , This is a preset bias term; S54, score the tissue recovery. With the probability of recurrence To integrate, if the risk of recurrence is high. If the risk exceeds the preset risk threshold, the corresponding area will be highlighted in pseudo-color to generate a comprehensive rehabilitation assessment report that includes quantitative indicators, historical trend line charts, and early warning image slices.

[0013] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This CT image adaptive enhancement breast tumor recovery assessment system eliminates radial metallic artifacts by introducing a dual-domain collaborative correction mechanism of projection domain adaptive interpolation and image domain GAN network. It overcomes the edge smoothing distortion easily caused by traditional single projection domain repair, accurately restores the real anatomical structure of the area adjacent to the implant, and by mining the texture information of local radiomics, combined with an adaptive nonlinear grayscale mapping algorithm, it customizes the dynamic range of the soft tissue window, amplifies the subtle differences between postoperative fibrotic scars, edema and micro-recurrent tumors, and links historical images to construct a spatiotemporal evolution model, transforming static image observation into dynamic trend tracking, providing a scientific assessment basis for clinical follow-up. Attached Figure Description

[0014] Figure 1 A schematic diagram of the overall structure of the present invention is shown; Figure 2 A schematic diagram of the overall method flow of the present invention is shown. Detailed Implementation

[0015] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example: like Figure 1-2 As shown, a CT image adaptive enhancement breast tumor recovery assessment system includes an image data acquisition module, an adaptive enhancement module, a feature perception module, and a recovery level quantitative assessment module, wherein: The image data acquisition module is used to acquire the original projection data of the patient's postoperative breast region and the initial CT images of preliminary reconstruction, and send them to the adaptive enhancement module. The adaptive enhancement module includes a high-density material segmentation unit and a dual-domain artifact correction unit, wherein: The high-density material segmentation unit is used to acquire initial CT images, identify and extract prior information on the spatial location of the high-density material distribution area based on the initial CT images, and the high-density material includes silicone prostheses, tissue expanders and titanium clips. The specific process for delineating high-density material distribution areas is as follows: S11. Obtain the initial CT image of preliminary reconstruction, traverse the CT value of each voxel point in the initial CT image, obtain the preset material density range and perform multi-level threshold segmentation, and divide the first candidate region and the second candidate region in the initial CT image. Specifically, for silicone implants and tissue expanders, a first threshold range (preferably 100 HU to 350 HU) is set to divide the first candidate region; For the titanium clip, a second threshold range (preferably greater than or equal to 2000 HU) is set to divide the second candidate region; S12. Perform three-dimensional morphological opening and closing operations on the first candidate region and the second candidate region respectively, remove isolated bright noise points with a volume smaller than the first set threshold, and smooth the three-dimensional edge structure of the high-density candidate region.

[0017] The specific process of obtaining prior information about spatial location is as follows: S21. The first candidate region and the second candidate region are marked as connected components, and the geometric feature parameters of each connected component are calculated. The geometric feature parameters include the volume, surface area and shape anisotropy index of the connected component. S22. Compare the calculated geometric feature parameters with the preset chest bone atlas feature library. If the shape anisotropy index of a connected region is a slender tubular structure and the volume conforms to the human bone characteristics, then the connected region is removed from the first candidate region and the second candidate region, thereby obtaining the accurate silicone implant distribution area and the titanium clip distribution area. S23. The silicone prosthesis distribution area and the titanium clip distribution area are binarized to generate a three-dimensional spatial position mask of the high-density material. Preset CT scan geometric parameters are obtained. A forward projection operation is performed on the three-dimensional spatial position mask to map it from the three-dimensional image domain to the two-dimensional sinusoidal domain, generating a corresponding projection trajectory mask. The projection trajectory mask is the spatial position prior information, used to indicate the range of projection data that needs to be discarded and interpolated in the original projection data due to the attenuation of rays from the high-density material.

[0018] The dual-domain artifact correction unit utilizes the prior information of spatial location and divides the original projection domain and image domain on the initial CT image according to the original projection data. In the original projection domain, it performs adaptive interpolation repair on the projection trajectory obscured by high-density material. At the same time, in the image domain, it uses a generative adversarial network to deeply suppress residual artifacts. After obtaining the repaired projection data, it performs secondary reconstruction to generate an artifact-free basic image and sends it to the feature perception module. The specific process of adaptive interpolation repair is as follows: The original projection data is acquired, and forward projection is performed using a preset CT scan geometry matrix to map the spatial location prior information into the original projection data, thereby generating a projection trajectory mask corresponding to the rays blocked by high-density material. Using the projection trajectory mask as an index, the original projection data is divided into a reliable projection area and a damaged projection area. For data points in the damaged projection area, the data in the reliable projection area in the neighborhood of the data point is used to perform adaptive interpolation repair using a normalized metal artifact interpolation algorithm to obtain the initially repaired projection data.

[0019] Specifically, the initial CT images are first used to generate a priori images without high-density materials, which are then normalized. After completing one-dimensional linear interpolation or spline interpolation, inverse normalization is performed to restore the images and obtain the projection data for the initial physical repair. The specific process for generating the artifact-free base image is as follows: S31. Construct a generative adversarial network model containing a generator and a discriminator. The generator adopts an encoder-decoder architecture. Its input is the intermediate transition image obtained by the first filtering back projection reconstruction of the initially repaired projection data. The generator extracts the deep features of residual water ripple artifacts and soft tissue edges through multi-layer convolution and outputs the predicted artifact distribution map. The intermediate transition image and the predicted artifact distribution map are subjected to difference operation to obtain the initial artifact-free image. S32. The discriminator is used to determine whether the texture distribution of the preliminary artifact removal image conforms to the real breast soft tissue anatomical structure. The generator network weights are continuously optimized through adversarial training until a high-fidelity image domain artifact removal feature matrix is ​​output. S33. The preliminary repaired projection data and the image domain artifact removal feature matrix are jointly iterated. During the secondary image reconstruction, the image domain artifact removal feature matrix is ​​used as a regularization constraint term to perform iterative convergence reconstruction on the repaired projection data, and finally generate an artifact-free basic image that effectively eliminates high-density implant radial artifacts and secondary stripe interference, and completely preserves the surrounding soft tissue anatomical structure.

[0020] The feature perception module is used to receive the artifact-removed base image, extract tissue texture information based on local radiomics features, use an adaptive nonlinear grayscale mapping algorithm to stretch the grayscale dynamic range within the soft tissue window, enhance the density and texture differences between fibrotic scars, edema tissue and active tumor cell clusters, and generate a high-contrast enhanced image to be sent to the recovery degree quantification assessment module. The specific process for generating high-contrast enhanced images is as follows: S41. Obtain the artifact-free base image. Based on a preset breast soft tissue CT threshold window (preferably 20 HU to 60 HU), segment the soft tissue region of interest containing the tumor bed and its edge. Calculate the three-dimensional gray-level co-occurrence matrix and local binary mode of each voxel block within the region of interest, thereby obtaining texture parameters characterizing the tissue microstructure. The texture parameters include local information entropy, angular second moment, and gray-level difference contrast. S42. Integrate the texture parameters to construct a tissue heterogeneity index. The tissue heterogeneity index is positively correlated with local information entropy and gray-level difference contrast, and negatively correlated with the second moment of angle. It is used to quantitatively distinguish between uniform edema, scar tissue and non-uniform suspected active tumor masses. S43, The tissue heterogeneity index As an adjustment parameter, an adaptive nonlinear gray-scale mapping model is constructed using an improved adaptive Sigmoid function, and the specific formula is as follows: ; in, The initial CT grayscale value of the voxel; The target grayscale value after mapping; Display the maximum value of the dynamic range for the target; This is the center offset of the soft tissue window; This is the gain coefficient; The preset heterogeneity index judgment interval is obtained. When the heterogeneity index of a local area is extremely high, it is judged as a suspected tumor hyperplasia area. At this time, the slope of the mapping curve near the center offset increases sharply, thereby non-linearly and sharply stretching the subtle differences in CT values. When the heterogeneity index of a local area is extremely low, it is identified as a homogeneous scar or edema area, and a gentler mapping slope is used. In this way, the difference between scars, edema and active tumors is maximized in the numerical domain. S44. Input the low-frequency background layer of the target tissue image and the artifact-free base image into the Laplacian pyramid for multi-scale smooth fusion, and output a high-contrast enhanced image with clear soft tissue structure and sharp separation of scar and recurrent lesion boundaries.

[0021] The quantitative assessment module for recovery level is used to acquire the high-contrast enhanced images to calculate the characteristic index vector of the tumor bed and marginal area after breast tumor surgery. It combines the patient's preoperative historical image sequence to construct a spatiotemporal evolution model and output a recovery assessment report that includes tissue recovery score and recurrence risk probability.

[0022] The specific process for generating a rehabilitation assessment report is as follows: S51. Acquire high-contrast enhanced images, divide the tumor bed core area and the edge extension area at a specific distance based on image threshold and morphological algorithm, extract objective physical features in the edge extension area, and calculate feature index vector. The feature index vector includes volume change index reflecting tissue proliferation / absorption, boundary sharpness index reflecting interface stability, and three-dimensional gray-level gradient variance reflecting microvascular density. S52. Obtain the preoperative and postoperative historical image sequences from the patient's electronic medical record. Based on the B-spline non-rigid registration algorithm, using the patient's stable sternum and rib contours as strong constraint reference points, elastically register and deform the historical image sequences with the current high-contrast enhanced images. Stack the feature index vectors of the registered images at each time point in chronological order to construct a multi-dimensional spatiotemporal feature sequence map X = { , , ..., }, where t is the current time node; S53. Input the spatiotemporal feature sequence map into a pre-trained spatiotemporal evolution model. The spatiotemporal evolution model uses a convolutional long short-term memory network. Through multi-layer forward propagation of the convolutional long short-term memory network, a deep spatiotemporal fusion feature matrix representing the dynamic evolution law of the organization is extracted. Deep spatiotemporal fusion feature matrix The input is fed into the terminal double-branch fully connected layer of the convolutional long short-term memory network. The terminal double-branch fully connected layer includes a first branch and a second branch: The first branch performs a regression task, calculating the current tissue recovery score using the logistic activation function. The formula is as follows: ; The second branch performs the classification probability prediction task, calculating the relapse risk probability using a normalized exponential function. The formula is as follows: ; in, , For the preset weight matrix , This is a preset bias term; S54, score the tissue recovery. (A higher tissue recovery score indicates more mature edema reduction and benign scarring) and a higher probability of recurrence. To integrate, if the risk of recurrence is high. If the risk exceeds the preset risk threshold, the corresponding area will be highlighted in pseudo-color to generate a comprehensive rehabilitation assessment report that includes quantitative indicators, historical trend line charts, and early warning image slices.

[0023] This invention eliminates radial metal artifacts by introducing a dual-domain collaborative correction mechanism of adaptive interpolation in the projection domain and a GAN network in the image domain. It overcomes the edge smoothing distortion that is easily caused by traditional single projection domain repair, accurately restores the real anatomical structure of the area adjacent to the implant, and by mining the texture information of local radiomics, combined with an adaptive nonlinear grayscale mapping algorithm, it customizes the dynamic range of the soft tissue window, amplifies the subtle differences between postoperative fibrotic scars, edema and micro-recurrent tumors, and at the same time links historical images to construct a spatiotemporal evolution model, transforming static image observation into dynamic trend tracking, and providing a scientific evaluation basis for clinical follow-up.

[0024] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0025] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A breast tumor recovery assessment system based on adaptive enhancement of CT images, characterized in that, It includes an image data acquisition module, an adaptive enhancement module, a feature perception module, and a rehabilitation degree quantitative assessment module, among which: The image data acquisition module is used to acquire the original projection data of the patient's postoperative breast region and the initial CT images of preliminary reconstruction, and send them to the adaptive enhancement module. The adaptive enhancement module includes a high-density material segmentation unit and a dual-domain artifact correction unit, wherein: The high-density material segmentation unit is used to acquire initial CT images, identify and extract prior information on the spatial location of the high-density material distribution area based on the initial CT images, and the high-density material includes silicone prostheses, tissue expanders and titanium clips. The dual-domain artifact correction unit utilizes the prior information of spatial location and divides the original projection domain and image domain on the initial CT image according to the original projection data. In the original projection domain, it performs adaptive interpolation repair on the projection trajectory obscured by high-density material. At the same time, in the image domain, it uses a generative adversarial network to deeply suppress residual artifacts. After obtaining the repaired projection data, it performs secondary reconstruction to generate an artifact-free basic image and sends it to the feature perception module. The feature perception module is used to receive the artifact-removed base image, extract tissue texture information based on local radiomics features, use an adaptive nonlinear grayscale mapping algorithm to stretch the grayscale dynamic range within the soft tissue window, enhance the density and texture differences between fibrotic scars, edema tissue and active tumor cell clusters, and generate a high-contrast enhanced image to be sent to the recovery degree quantification assessment module. The recovery level quantitative assessment module is used to acquire the high-contrast enhanced images to calculate the characteristic index vector of the tumor bed and marginal area after breast tumor surgery. It combines the patient's preoperative historical image sequence to construct a spatiotemporal evolution model and output a recovery assessment report that includes tissue recovery score and recurrence risk probability.

2. The CT image adaptive enhancement breast tumor recovery assessment system according to claim 1, characterized in that, The specific process for delineating high-density material distribution areas is as follows: S11. Obtain the initial CT image of preliminary reconstruction, traverse the CT value of each voxel point in the initial CT image, obtain the preset material density range and perform multi-level threshold segmentation, and divide the first candidate region and the second candidate region in the initial CT image. Specifically, for silicone implants and tissue expanders, a first threshold interval is set to divide the first candidate region; For the titanium clip, a second threshold interval is set to divide the second candidate region; S12. Perform three-dimensional morphological opening and closing operations on the first candidate region and the second candidate region respectively, remove isolated bright noise points with a volume smaller than the first set threshold, and smooth the three-dimensional edge structure of the high-density candidate region.

3. The CT image adaptive enhancement breast tumor recovery assessment system according to claim 1, characterized in that, The specific process of identifying and extracting high-density material distribution areas based on initial CT images is as follows: S21. The first candidate region and the second candidate region are marked as connected components, and the geometric feature parameters of each connected component are calculated. The geometric feature parameters include the volume, surface area and shape anisotropy index of the connected component. S22. Compare the calculated geometric feature parameters with the preset chest bone atlas feature library. If the shape anisotropy index of a connected region is a slender tubular structure and the volume conforms to the characteristics of human bones, then the connected region is removed from the first candidate region and the second candidate region, thereby obtaining the accurate silicone implant distribution area and the titanium clip distribution area.

4. The CT image adaptive enhancement breast tumor recovery assessment system according to claim 3, characterized in that, The specific process of obtaining prior information about spatial location is as follows: The distribution areas of the silicone prosthesis and the titanium clip are binarized to generate a three-dimensional spatial position mask of the high-density material. Preset CT scan geometric parameters are obtained, and a forward projection operation is performed on the three-dimensional spatial position mask to map it from the three-dimensional image domain to the two-dimensional sinusoidal domain, generating a corresponding projection trajectory mask. The projection trajectory mask is the spatial position prior information, used to indicate the range of projection data that needs to be discarded and interpolated in the original projection data due to the attenuation of rays from the high-density material.

5. The CT image adaptive enhancement breast tumor recovery assessment system according to claim 1, characterized in that, The specific process of adaptive interpolation repair is as follows: The original projection data is acquired, and forward projection is performed using a preset CT scan geometry matrix to map the spatial location prior information into the original projection data, thereby generating a projection trajectory mask corresponding to the rays blocked by high-density material. Using the projection trajectory mask as an index, the original projection data is divided into a reliable projection area and a damaged projection area. For data points in the damaged projection area, the data in the reliable projection area in the neighborhood of the data point is used to perform adaptive interpolation repair using a normalized metal artifact interpolation algorithm to obtain the initially repaired projection data.

6. The CT image adaptive enhancement breast tumor recovery assessment system according to claim 1, characterized in that, The specific process for generating the artifact-free base image is as follows: S31. Construct a generative adversarial network model containing a generator and a discriminator. The generator adopts an encoder-decoder architecture. Its input is the intermediate transition image obtained by the first filtering back projection reconstruction of the initially repaired projection data. The generator extracts the deep features of residual water ripple artifacts and soft tissue edges through multi-layer convolution and outputs the predicted artifact distribution map. The intermediate transition image and the predicted artifact distribution map are subjected to difference operation to obtain the initial artifact-free image. S32. The discriminator is used to determine whether the texture distribution of the preliminary artifact removal image conforms to the real breast soft tissue anatomical structure. The generator network weights are continuously optimized through adversarial training until a high-fidelity image domain artifact removal feature matrix is ​​output. S33. The initially repaired projection data and the image domain artifact removal feature matrix are jointly iterated. During the secondary image reconstruction process, the image domain artifact removal feature matrix is ​​used as a regularization constraint term to perform iterative convergence reconstruction on the repaired projection data, and finally generate the artifact removal base image.

7. The CT image adaptive enhancement breast tumor recovery assessment system according to claim 1, characterized in that, The specific process for generating high-contrast enhanced images is as follows: S41. Obtain the artifact-free basic image. Based on the preset breast soft tissue CT threshold window, segment the soft tissue region of interest containing the tumor bed and its edge. Calculate the three-dimensional gray-level co-occurrence matrix and local binary mode of each voxel block within the region of interest, thereby obtaining texture parameters characterizing the tissue microstructure. The texture parameters include local information entropy, second angular moment, and gray-level difference contrast. S42. Integrate the texture parameters to construct a tissue heterogeneity index. The tissue heterogeneity index is positively correlated with local information entropy and gray-level difference contrast, and negatively correlated with the second moment of angle. It is used to quantitatively distinguish between uniform edema, scar tissue and non-uniform suspected active tumor masses. S43, The tissue heterogeneity index As an adjustment parameter, an adaptive nonlinear gray-scale mapping model is constructed using an improved adaptive Sigmoid function, and the specific formula is as follows: ； in, The initial CT grayscale value of the voxel; The target grayscale value after mapping; Display the maximum value of the dynamic range for the target; This is the center offset of the soft tissue window; This is the gain coefficient; S44. Input the low-frequency background layer of the target tissue image and the artifact-removed base image into the Laplacian pyramid for multi-scale smooth fusion, and output a high-contrast enhanced image.

8. The CT image adaptive enhancement breast tumor recovery assessment system according to claim 1, characterized in that, The specific process for calculating the characteristic index vector of the tumor bed and marginal region after breast tumor surgery is as follows: High-contrast enhanced images are acquired, and the tumor bed core area and the edge extension area extending to a specific distance are divided based on image threshold and morphological algorithm. Objective physical features within the edge extension area are extracted, and feature index vectors are calculated. The feature index vectors include a volume change index reflecting tissue proliferation / absorption, a boundary sharpness index reflecting interface stability, and a three-dimensional gray-level gradient variance reflecting microvascular density.

9. A CT image adaptive enhancement breast tumor recovery assessment system according to claim 1, characterized in that, The specific process of constructing the spatiotemporal evolution model is as follows: S51. Obtain the preoperative and postoperative historical image sequences from the patient's electronic medical record. Based on the non-rigid registration algorithm of B-spline, take the patient's stable sternum and rib contours as strong constraint reference points, perform elastic registration deformation on the historical image sequences and the current high-contrast enhanced images, and stack the feature index vectors of the registered images at each time node in chronological order to construct a multi-dimensional spatiotemporal feature sequence map. S52. The spatiotemporal feature sequence map is input into a pre-trained spatiotemporal evolution model. The spatiotemporal evolution model uses a convolutional long short-term memory network. Through multi-layer forward propagation of the convolutional long short-term memory network, a deep spatiotemporal fusion feature matrix representing the dynamic evolution law of the organization is extracted. The deep spatiotemporal fusion feature matrix is ​​input into the terminal double-branch fully connected layer of the convolutional long short-term memory network. The terminal double-branch fully connected layer includes a first branch and a second branch. The first branch performs a regression task, calculating the current tissue recovery score using the logistic activation function. ; The second branch performs the classification probability prediction task, calculating the relapse risk probability using a normalized exponential function. .

10. A CT image adaptive enhancement breast tumor recovery assessment system according to claim 9, characterized in that, The specific process for generating a rehabilitation assessment report is as follows: The tissue recovery score With the probability of recurrence To integrate, if the risk of recurrence is high. If the risk exceeds the preset risk threshold, the corresponding area will be highlighted in pseudo-color to generate a comprehensive rehabilitation assessment report that includes quantitative indicators, historical trend line charts, and early warning image slices.

Images