Artificial intelligence-based medical image quality enhancement and artifact correction method and system

Abstract

The application relates to the technical field of image analysis, and provides a medical image quality enhancement and artifact correction method and system based on artificial intelligence, which solves the problem that when multiple physical cause artifacts are jointly processed, it is difficult to balance the correction effectiveness and structure fidelity. The method comprises the following steps: analyzing obtained constraints to obtain an intensity weight map and a continuity guide signal; performing layered feature extraction and weighted modulation on an obtained original medical image to obtain multi-scale features; constructing an intermediate image as a first medical image; identifying first and second artifact regions in the original medical image to generate an artifact response map, suppressing strip-shaped artifacts and volume effect artifacts of the original medical image to obtain a target medical image, repairing a tissue boundary region, and outputting a second medical image; and fusing the first and second medical images to output a final medical image. The technical scheme provided by the application not only improves the overall signal-to-noise ratio of the image, but also enhances the visibility of key anatomical details.

Description

Technical Field

[0001] This application relates to the field of image analysis technology, and in particular to a method and system for medical image quality enhancement and artifact correction based on artificial intelligence. Background Technology

[0002] With the widespread application of low-dose CT scans in clinical screening and diagnosis, how to generate high-quality, low-artifact medical images based on limited scan data has become an urgent technical requirement. This not only requires algorithms to improve the overall signal-to-noise ratio of the image while enhancing the visibility of key anatomical details, but also to accurately correct for issues such as strip artifacts introduced by low-dose conditions and blurred tissue boundaries caused by partial volume effects.

[0003] Currently, existing technical solutions employ end-to-end deep neural networks. These networks are trained using a large number of pairs of low-quality and high-quality CT image data to learn a direct mapping from low-dose CT images to standard-dose CT images. Through a single model architecture, they simultaneously enhance the overall image quality and suppress common artifacts.

[0004] However, this technical solution has significant drawbacks. First, its model processing lacks explicit utilization of specific scanning parameters and anatomical site-specific information, resulting in a lack of targeted correction capabilities for artifact types with different physical causes generated by different scanning protocols and anatomical structures. Second, due to the use of a single processing path, the model is prone to compromise between enhancing details and suppressing artifacts during the optimization process. It is difficult to maintain the integrity of fine anatomical structures while suppressing specific scanning artifacts and correcting artifacts in specific anatomical regions, especially at low-contrast soft tissue boundaries, which can easily lead to loss of details or texture distortion. Summary of the Invention

[0005] This application provides a method and system for medical image quality enhancement and artifact correction based on artificial intelligence, in order to solve the problem in the prior art that it is difficult to balance the effectiveness of correction and structural fidelity when jointly processing artifacts of multiple physical causes.

[0006] Firstly, this application provides an artificial intelligence-based method for medical image quality enhancement and artifact correction, including:

[0007] The original medical image of the target object and the constraints matching the original medical image are obtained, and the constraints are analyzed using a pre-built image processing model to obtain the intensity weight map and continuity guidance signal of the constraints.

[0008] Based on the intensity weight map, the original medical image is subjected to hierarchical feature extraction and weighted modulation to obtain multi-scale features of the original medical image;

[0009] Based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image and the background brightness component of the original medical image, an intermediate image is constructed, and the texture smoothing transition of the intermediate image is performed using the continuity guidance signal to output the first medical image.

[0010] Based on the constraints, identify the first and second type artifact regions in the original medical image and generate corresponding artifact shadow maps;

[0011] Based on the artifact suppression map, the strip artifacts and volume effect artifacts of the original medical image are suppressed to obtain the target medical image. Based on the structural integrity information in the constraints, the texture consistency of the tissue boundary region in the target medical image is restored, and the second medical image is output.

[0012] The first medical image and the second medical image are fused together to output the final medical image.

[0013] Optionally, the constraints are analyzed using a pre-built image processing model to obtain the intensity weight map and continuity guidance signal of the constraints, including:

[0014] The constraints are input into the condition parsing module of the image processing model;

[0015] The constraint intensity distribution information corresponding to different spatial scales in the constraint conditions is extracted through the first analytical branch in the condition analysis module.

[0016] The constraint intensity distribution information is scale-aligned and nonlinearly mapped to generate an intensity weight map that matches the pixel position of the original medical image.

[0017] The topological relationship information characterizing the continuity of the anatomical structure boundary is extracted from the constraint conditions through the second analytical branch in the condition analysis module.

[0018] Based on the topological relationship information, the expected values ​​of connection strength and transition smoothness between adjacent anatomical regions in the original medical image are calculated.

[0019] Based on the connection strength and the expected transition smoothness, a continuous guiding signal is generated to guide the smooth transition of local textures in the image.

[0020] Optionally, based on the intensity weight map, hierarchical feature extraction and weighted modulation are performed on the original medical image to obtain multi-scale features of the original medical image, including:

[0021] The original medical image is input into a preset feature extraction network, and the global contour features, local tissue features and microscopic detail features of the original medical image are extracted by the convolutional layer groups with different receptive fields in the feature extraction network.

[0022] Separate the first-scale weight distribution, the second-scale weight distribution, and the third-scale weight distribution corresponding to the global contour feature, the local tissue feature, and the microscopic detail feature from the intensity weight map;

[0023] The first scale weight distribution is multiplied element-wise with the global contour feature to obtain the weighted modulated global contour feature.

[0024] The second-scale weight distribution is multiplied element-wise with the local tissue features to obtain the weighted modulated local tissue features;

[0025] The third-scale weight distribution is multiplied element-wise with the micro-detail features to obtain the weighted modulated micro-detail features;

[0026] The weighted modulated global contour features, the weighted modulated local tissue features, and the weighted modulated microscopic detail features are spliced ​​together according to the feature channel dimension to form the multi-scale features of the original medical image.

[0027] Optionally, based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image, and the background brightness component of the original medical image, an intermediate image is constructed, and the texture smoothing transition of the intermediate image is performed using the continuity guidance signal to output a first medical image, including:

[0028] The weighted modulated microscopic detail features are separated from the multi-scale features as the detail features to be compensated;

[0029] Based on the consistency of gradient direction of each pixel in the weighted modulated local tissue features, a detail compensation function is constructed.

[0030] The detail features to be compensated are nonlinearly enhanced by a detail compensation function to generate enhanced detail features.

[0031] The detailed enhancement features are fused with the weighted modulated local tissue features to obtain the tissue detail fusion features.

[0032] The structural enhancement features are obtained by concatenating the fused organizational details with the weighted modulated global contour features.

[0033] The structural enhancement features are superimposed on the background brightness component of the original medical image at the pixel level to generate an initial intermediate image;

[0034] Extract the transition region in the initial intermediate image that corresponds to the anatomical boundary indicated by the continuous guiding signal;

[0035] Based on the smoothness expectation value in the continuous guidance signal, the gray values ​​of adjacent pixels in the transition region of the initial intermediate image are subjected to diffusion balance processing to obtain the intermediate image.

[0036] The intermediate image is subjected to brightness normalization processing to output the first medical image.

[0037] Secondly, this application provides an artificial intelligence-based medical image quality enhancement and artifact correction system, comprising:

[0038] The acquisition module is used to acquire the original medical image of the target object and the constraint conditions matching the original medical image, and to analyze the constraint conditions using a pre-built image processing model to obtain the intensity weight map and continuity guidance signal of the constraint conditions.

[0039] The extraction module is used to perform hierarchical feature extraction and weighted modulation on the original medical image based on the intensity weight map to obtain the multi-scale features of the original medical image;

[0040] The construction module is used to construct an intermediate image based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image and the background brightness component of the original medical image, and to perform texture smoothing transition processing on the intermediate image using the continuity guidance signal to output the first medical image.

[0041] The generation module is used to identify the first and second type artifact regions in the original medical image based on the constraints and generate corresponding artifact shadow maps.

[0042] The repair module is used to suppress the strip artifacts and volume effect artifacts of the original medical image according to the artifact effect map, obtain the target medical image, and perform texture consistency repair on the tissue boundary region in the target medical image based on the structural integrity information in the constraint conditions, and output the second medical image.

[0043] The fusion module is used to fuse the first medical image and the second medical image to output the final medical image.

[0044] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement an artificial intelligence-based medical image quality enhancement and artifact correction method as described in the first aspect above.

[0045] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements an artificial intelligence-based medical image quality enhancement and artifact correction method as described in the first aspect.

[0046] The beneficial effects of this application are:

[0047] This application introduces constraints that match the original medical image data, and based on the intensity weight map and continuity guidance signal parsed from these constraints, it achieves hierarchical extraction and weighted modulation of multi-scale features of the image. This method can effectively utilize the prior knowledge contained in the scanning protocol parameters and anatomical structure identification information. In the process of detail restoration, it constructs an intermediate image by combining the background brightness component with structural consistency enhancement and detail compensation algorithms, and uses the continuity guidance signal to ensure smooth texture transition. This improves the overall signal-to-noise ratio of the image while enhancing the visibility of key anatomical details, avoiding the problems of detail loss or texture distortion caused by conventional enhancement methods.

[0048] Furthermore, by precisely identifying the first and second type artifact regions caused by scanning protocol parameters and target anatomical structure characteristics through constraints, and generating targeted artifact response maps, the specific suppression of strip artifacts and volume effect artifacts is achieved. After artifact suppression, graded texture restoration processing is performed on tissue boundary areas based on structural integrity information. The restoration strategy is dynamically adjusted according to the matching degree between the reference texture sample and the restoration template, ensuring the consistency of the restoration area with the surrounding healthy tissue in texture direction and grayscale transition. This collaborative processing mechanism enables the final output medical image to maintain the integrity and natural coherence of fine anatomical structures while suppressing artifacts of various physical causes, solving the technical problem of existing technologies that are difficult to balance correction effectiveness and structural fidelity when jointly processing multiple types of artifacts.

[0049] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A flowchart of an artificial intelligence-based medical image quality enhancement and artifact correction method provided in this application is shown;

[0052] Figure 2 A schematic diagram of the structure of an artificial intelligence-based medical image quality enhancement and artifact correction system provided in this application is shown.

[0053] Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0054] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0055] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0056] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0057] Figure 1 This application provides a flowchart of an artificial intelligence-based medical image quality enhancement and artifact correction method, such as... Figure 1 As shown, the method includes:

[0058] Step 101: Obtain the original medical image of the target object and the constraint conditions matching the original medical image, and use the pre-built image processing model to analyze the constraint conditions to obtain the intensity weight map and continuity guidance signal of the constraint conditions.

[0059] Optionally, step 101 may specifically include:

[0060] Step 1011: Input the constraints into the condition parsing module in the image processing model.

[0061] Step 1012: Extract the constraint intensity distribution information corresponding to different spatial scales from the constraint conditions through the first analytical branch in the condition analysis module.

[0062] Step 1013: Scale alignment and nonlinear mapping are performed on the constraint intensity distribution information to generate an intensity weight map that matches the pixel position of the original medical image.

[0063] Step 1014: Extract the topological relationship information representing the continuity of the anatomical structure boundary from the constraint conditions through the second parsing branch in the condition parsing module.

[0064] Step 1015: Based on the topological relationship information, calculate the expected values ​​of connection strength and transition smoothness between adjacent anatomical regions in the original medical image.

[0065] Step 1016: Based on the connection strength and the expected transition smoothness value, generate a continuity guidance signal to guide the smooth transition of local textures in the image.

[0066] In this step, raw medical images refer to the initial image data directly acquired by medical imaging equipment without any processing. This data serves as the input basis for all subsequent image enhancement and correction operations and is obtained by scanning the examined area using scanning equipment such as CT and MRI.

[0067] Constraints refer to a set of prior information related to the original medical image, including the parameter settings of the scanning equipment and the anatomical structure identification of the scanned object. They are used to provide guidance rules for targeted processing of the image processing model and are generated by integrating scanning protocol parameters and anatomical structure identification information.

[0068] Pre-built image processing models refer to computer program models that have been trained using a large amount of medical image data and have specific image processing capabilities. They are used to automatically perform a series of complex operations from constraint analysis to image fusion, and are obtained through adversarial training on massive pairs of low-quality and high-quality medical image datasets.

[0069] An intensity weight map is a distribution map of weight values ​​corresponding to each pixel position in the original image, generated by analyzing constraints. It is used to indicate the enhancement or correction intensity that should be applied to different regions in subsequent processing.

[0070] Continuity guidance signal refers to guidance information generated by analyzing the boundary relationships of anatomical structures, used to describe the ideal transition between different tissue regions in an image, and to ensure that the processed image remains natural and smooth at tissue boundaries.

[0071] The condition parsing module is a dedicated component within the image processing model, specifically responsible for decomposing and analyzing the input constraints, and is integrated into the image processing model during the model design phase.

[0072] The first parsing branch is the part of the condition parsing module that focuses on extracting intensity distribution information related to spatial scale from the constraints. It is achieved by parsing the correspondence between scanning protocol parameters (such as layer thickness and tube voltage) and image scale.

[0073] The constraint intensity distribution information is the raw information about how the processing intensity should be distributed at different scales, which is initially extracted from the constraint conditions. It is obtained by analyzing the scanning protocol parameters and the scale characteristics of the anatomical structure through the first analytical branch.

[0074] The second parsing branch is the part of the condition parsing module that focuses on extracting topological relationship information related to the continuity of anatomical structure boundaries from the constraints. It is achieved by parsing the spatial relationships of tissue boundaries in the anatomical structure identification information.

[0075] Topological relationship information refers to data describing the spatial positional relationships such as adjacency and connection between different anatomical structures in an image, which is obtained by analyzing the tissue boundary connection relationships in standard anatomical atlases through the second analytical branch.

[0076] The expected values ​​of connection strength and transition smoothness are numerical targets calculated by analyzing topological relationship information. They quantify the degree of connection tightness and the smoothness of gray-level changes that adjacent anatomical regions should have. They are obtained by calculating the gradient correlation of the boundary pixels of adjacent anatomical regions and the expected gray-level change rate.

[0077] In this step, firstly, the constraints are input into the condition parsing module of the pre-built image processing model through a dual-branch parallel processing architecture.

[0078] Secondly, in the first parsing branch of the condition parsing module, in order to extract the constraint intensity distribution information corresponding to different spatial scales in the constraint conditions, a filtering algorithm based on scale space theory is adopted, such as constructing a Gaussian pyramid or using convolution kernels of different sizes, to actively analyze the influence of scanning protocol parameters on image features at different spatial scales, thereby obtaining multi-scale constraint intensity distribution information.

[0079] Next, scale alignment and nonlinear mapping are performed on these multi-scale constraint intensity distribution information: through bilinear interpolation or upsampling algorithms, the constraint intensity distribution information of all scales is uniformly upsampled to the same spatial resolution as the original medical image; then, through a preset nonlinear activation function, such as the Sigmoid function, the aligned intensity values ​​are nonlinearly mapped to model the complex relationship between physical parameters and expected processing intensity, and finally an intensity weight map matching the pixel position of the original medical image is generated.

[0080] Meanwhile, in the second parsing branch of the condition parsing module, in order to extract the topological relationship information representing the continuity of the anatomical structure boundary in the constraint conditions, a graph theory analysis algorithm is used to abstract the anatomical structure identification information into a graph structure of nodes and edges, and a graph traversal algorithm, such as depth-first search, is used to analyze the topological relationships such as adjacency and containment between different anatomical regions.

[0081] Then, based on the extracted topological relationship information, the connection strength and expected transition smoothness between adjacent anatomical regions in the original medical image are calculated by an algorithm based on gradient field analysis: on both sides of the identified tissue boundary, the consistency of the gradient magnitude and direction of the pixel gray value is calculated, where the connection strength is determined by the degree of coordination of the gradient directions on both sides of the boundary, and the expected transition smoothness is derived from the statistical distribution difference of gray values ​​in a certain area on both sides of the boundary.

[0082] Finally, based on the calculated connection strength and expected transition smoothness, a continuous guiding signal is generated to guide the smooth transition of local textures in the image by using a signal synthesis algorithm, such as modulating a basic Gaussian kernel function or smoothing filter based on these values ​​as parameters.

[0083] For example, in the radiology department of hospital A, a doctor performs a low-dose chest CT scan on patient B. First, the doctor acquires the raw medical image data obtained from the scan and simultaneously records parameters such as tube voltage and slice thickness used in the scan. At the same time, the doctor retrieves the patient's standard anatomical atlas of the chest from the PACS system as anatomical structure identification information. These together constitute the constraints. Second, the doctor calls a pre-built image processing model trained on tens of thousands of high-quality chest CT images. Then, the conditional parsing module within the model initiates a two-branch parsing process: The first branch analyzes the parameters such as the scan slice thickness and concludes that the small nodule areas in the lungs require high-intensity local enhancement, while the mediastinal large vessel areas require smooth global enhancement. After scale alignment and nonlinear mapping, an intensity weight map is generated, in which the weight values ​​of the lung fields and suspicious nodules are significantly higher. At the same time, the second branch identifies the boundary topological relationships between key structures such as the lungs, heart, and blood vessels based on the chest anatomical atlas. It calculates the required connection strength and smooth transition expectation values ​​at the junctions of various tissues, such as the junction between the edge of the heart shadow and the lung tissue. Finally, a continuous guiding signal is generated to guide subsequent processing to achieve fine smoothing at key boundaries and avoid artifacts.

[0084] Step 102: Based on the intensity weight map, perform hierarchical feature extraction and weighted modulation on the original medical image to obtain the multi-scale features of the original medical image.

[0085] Optionally, step 102 may specifically include:

[0086] Step 1021: Input the original medical image into a preset feature extraction network, and extract the global contour features, local tissue features and microscopic detail features of the original medical image through convolutional layers with different receptive fields in the feature extraction network.

[0087] Step 1022: Separate the first-scale weight distribution, the second-scale weight distribution, and the third-scale weight distribution from the intensity weight map, which correspond to the global contour features, the local tissue features, and the microscopic detail features, respectively.

[0088] Step 1023: Multiply the first scale weight distribution with the global contour feature element by element to obtain the weighted modulated global contour feature.

[0089] Step 1024: Multiply the second scale weight distribution element-wise with the local tissue features to obtain the weighted modulated local tissue features.

[0090] Step 1025: Multiply the third-scale weight distribution with the micro-detail features element-wise to obtain the weighted modulated micro-detail features.

[0091] Step 1026: The weighted modulated global contour features, the weighted modulated local tissue features, and the weighted modulated microscopic detail features are spliced ​​together according to the feature channel dimension to form the multi-scale features of the original medical image.

[0092] In this step, multi-scale features refer to the set of features extracted from the original medical image that contains different levels of detail information, used to comprehensively describe the image information from the overall outline to local details.

[0093] A pre-designed feature extraction network refers to a pre-designed neural network structure containing multiple convolutional layers, used to automatically extract meaningful feature information from input images. It is trained on convolutional kernels of different sizes.

[0094] Global contour features refer to features that describe the overall shape and boundary information of the main tissue structures in an image. They are used to grasp the macroscopic anatomical structure of an image and are extracted through the convolutional layer group with the largest receptive field in the feature extraction network.

[0095] Local tissue features refer to the features that describe the texture and morphology of tissues in specific regions of an image. They are used to identify the internal structure of organs and are extracted through convolutional layers such as the receptive field in a feature extraction network.

[0096] Microscopic detail features refer to features that describe fine structures in an image, such as the edges of blood vessels and tiny lesions. They are used to enhance image resolution and are extracted by the convolutional layer group with the smallest receptive field in the feature extraction network.

[0097] The first-scale weight distribution refers to the weight value distribution map corresponding to the global contour features, which is used to adjust the contribution of the global contour features.

[0098] The second-scale weight distribution refers to the weight value distribution map corresponding to local organizational features, which is used to adjust the contribution of local organizational features.

[0099] The third-scale weight distribution refers to the weight value distribution map corresponding to the micro-detail features, which is used to adjust the contribution of the micro-detail features.

[0100] Weighted modulated global contour features refer to global contour features after weight adjustment, which are used to highlight important overall structural information.

[0101] Weighted modulation of local tissue features refers to local tissue features that have been adjusted by weights, which are used to enhance local information in key areas.

[0102] Weighted modulated micro-detail features refer to micro-detail features that have been adjusted by weights, and are used to enhance diagnostically valuable detailed information.

[0103] In this step, the original medical image is first input into a pre-defined feature extraction network using a multi-branch parallel convolutional structure. Each branch uses a different sized convolutional kernel to simulate different visual fields, thereby simultaneously extracting multi-scale features. Specifically, layers with large-sized convolutional kernels are used to capture the global contour features of the original medical image, layers with medium-sized convolutional kernels are used to extract local tissue features, and layers with small-sized convolutional kernels or depthwise separable convolutions are used to focus on the microscopic details of the original medical image.

[0104] Secondly, by performing scale-specific bilinear interpolation on the intensity weight map, and based on the spatial resolution of the global contour features, the local tissue features, and the microscopic detail features, the intensity weight map is precisely resampled into three weight distribution maps that perfectly match its size, namely the first-scale weight distribution, the second-scale weight distribution, and the third-scale weight distribution.

[0105] Next, the first-scale weight distribution is fused with the global contour features using an element-wise multiplication algorithm to generate a weighted modulated global contour feature. Similarly, the same algorithm is applied to multiply the second-scale weight distribution with the local tissue features to obtain a weighted modulated local tissue feature, and the third-scale weight distribution is multiplied with the microscopic detail features to obtain a weighted modulated microscopic detail feature.

[0106] Finally, through a channel stitching algorithm, the weighted modulated global contour features, the weighted modulated local tissue features, and the weighted modulated microscopic detail features are connected along the feature channel dimension, integrating the enhancement information at different scales to form the multi-scale features of the original medical image.

[0107] For example, following the specific implementation of the previous step, firstly, after generating the intensity weight map, the low-dose chest CT image of patient B, i.e., the original medical image, is input into the pre-trained feature extraction network; secondly, the large convolutional layer of the feature extraction network extracts the overall contour of the thoracic cavity as global contour features, the medium convolutional layer extracts the texture of the lung lobes, mediastinum, and other regions as local tissue features, and the small convolutional layer captures the edges of small lung nodules as microscopic detail features; then, the weight distribution corresponding to the three features is separated from the intensity weight map: the weight of the overall contour region is low, the weight of the lung field region is medium, and the weight of the suspicious nodule region is the highest; then, after element-wise multiplication modulation, the nodule detail features are significantly enhanced, while the contour features of the large blood vessel region are appropriately suppressed; finally, these modulated features are stitched together to form a comprehensive feature map containing multi-scale information, providing a rich feature foundation for subsequent processing.

[0108] Step 103: Based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image and the background brightness component of the original medical image, an intermediate image is constructed, and the texture smoothing transition of the intermediate image is performed using the continuity guidance signal to output the first medical image.

[0109] Optionally, step 103 may specifically include:

[0110] Step 1031: Separate the weighted modulated microscopic detail features from the multi-scale features as the detail features to be compensated.

[0111] Step 1032: Construct a detail compensation function based on the consistency of gradient direction of each pixel in the weighted modulated local tissue features.

[0112] Step 1033: The detail features to be compensated are nonlinearly enhanced using a detail compensation function to generate enhanced detail features.

[0113] Step 1034: The detail enhancement features are fused with the weighted modulated local tissue features to obtain the tissue detail fusion features.

[0114] Step 1035: The organizational detail fusion feature is combined with the weighted modulated global contour feature to obtain the structural enhancement feature.

[0115] Step 1036: The structural enhancement features are superimposed on the background brightness component of the original medical image at the pixel level to generate an initial intermediate image.

[0116] Step 1037: Extract the transition region in the initial intermediate image that corresponds to the anatomical boundary indicated by the continuity guidance signal.

[0117] Step 1038: Based on the smoothness expectation value in the continuity guidance signal, perform diffusion balance processing on the gray values ​​of adjacent pixels in the transition region of the initial intermediate image to obtain the intermediate image.

[0118] Step 1039: Perform brightness normalization processing on the intermediate image and output the first medical image.

[0119] In this step, the structural consistency enhancement and detail compensation algorithm refers to a set of combined techniques for improving the quality of medical images. This method enhances the clarity of image details while ensuring the natural continuity and integrity of anatomical structures. It is used to generate high-quality first medical images from raw medical images by coordinating nonlinear enhancement of detail features and constrained smoothing of structural features.

[0120] The background luminance component refers to the low-frequency information that represents the overall brightness and darkness of the image, separated from the original medical image. It is used to maintain the overall brightness consistency of the processed image and is extracted from the original image through high-pass filtering or image decomposition techniques.

[0121] Intermediate images refer to transitional image data generated during processing that has not yet undergone final optimization. They are used as input for subsequent texture smoothing and brightness normalization by superimposing structural enhancement features with the background brightness component.

[0122] The first medical image refers to the final output image after processing by the quality enhancement path, which is used to fuse with the second medical image output by the artifact correction path. It is obtained after completing all detail compensation, feature fusion and texture smoothing processes.

[0123] The details to be compensated refer to the micro-details selected from the multi-scale features that need to be further enhanced. They are used to improve the detail clarity of the image. They are obtained by separating the weighted and modulated micro-details from the multi-scale features.

[0124] Gradient direction consistency refers to a quantitative indicator that describes the consistency of pixel gray-level change direction within a local area of ​​an image. It is used to judge the rationality and intensity of detail enhancement and is obtained by calculating the standard deviation or consistency measure of pixel gradient direction within a local window.

[0125] The detail compensation function is a mathematical function used to non-linearly enhance detailed features, adaptively enhancing important details while suppressing noise.

[0126] Detail enhancement features refer to enhanced detail information after processing by a detail compensation function, used to improve the visibility of details in an image. They are obtained by applying the detail compensation function to the detail features to be compensated.

[0127] Organizational detail fusion features refer to the comprehensive features that combine enhanced details with local organizational features, used to maintain the coordination between details and organizational structure.

[0128] Structural enhancement features refer to complete feature representations that include global contours and local details, used to construct the enhanced image structure.

[0129] The initial intermediate image refers to the preliminary enhanced image that has not yet undergone texture smoothing, and is used for subsequent transition region optimization.

[0130] First, the weighted modulated microscopic detail features are separated from the multi-scale features of the original medical image using a feature channel selection algorithm, and these are used as the detail features to be compensated.

[0131] Secondly, based on the degree of gradient direction consistency of each pixel in the weighted modulated local tissue features, a detail compensation function is constructed. Specifically, the degree of gradient direction consistency of all pixels in the neighborhood around each pixel is calculated by a local gradient direction consistency analysis algorithm, and a nonlinear detail compensation function is constructed based on this degree map by lookup table mapping or piecewise function fitting algorithm.

[0132] Next, using the pointwise function mapping algorithm, the detail compensation function is used to nonlinearly enhance the detail features to be compensated, generating enhanced detail features.

[0133] Then, the detail enhancement features are fused with the weighted modulated local tissue features using an element-wise addition algorithm to obtain the tissue detail fusion features.

[0134] Subsequently, the tissue detail fusion features and the weighted modulated global contour features are concatenated using a channel stitching algorithm to obtain structural enhancement features.

[0135] Subsequently, the structural enhancement features are superimposed with the background brightness component of the original medical image at the pixel level using a pixel-by-pixel addition algorithm to generate an initial intermediate image.

[0136] Next, using image segmentation or edge detection algorithms, and combining the prior knowledge of anatomical boundaries encoded in the continuous guidance signal, the transition region corresponding to the anatomical boundary indicated by the continuous guidance signal in the initial intermediate image is extracted.

[0137] Finally, based on the expected smoothness value in the continuous guidance signal, the gray values ​​of adjacent pixels in the transition region of the initial intermediate image are diffused and balanced using an anisotropic diffusion algorithm to obtain an intermediate image, which is then output as the first medical image.

[0138] For example, following the specific implementation of the previous step, firstly, after obtaining a comprehensive feature map containing multi-scale information, the detail features representing the fine structures such as the edges of small lung nodules are separated from it to be compensated; secondly, the consistency of the gradient direction of each point in the lung tissue feature map is analyzed, and it is found that the gradient direction is highly consistent in the nodule edge region, while the direction is more random in the normal lung texture region. Based on this, a detail compensation function is constructed to focus on enhancing the nodule edge, generating detail enhancement features; then, the enhanced detail features are fused with the lung lobe texture features, and then stitched with the overall chest cavity contour features to form structural enhancement features; then, the structural enhancement features are superimposed with the overall brightness component of the original CT image to obtain the initial intermediate image; subsequently, based on the generated continuous guiding signal, key transition areas such as the cardiac shadow edge are located, and pixel grayscale diffusion balancing processing is performed in these areas to eliminate possible artificial processing traces; finally, after adjusting the overall brightness, a clear and natural first medical image is output.

[0139] Step 104: Identify the first and second type artifact regions in the original medical image based on the constraints and generate corresponding artifact reflection maps.

[0140] Optionally, step 104 may specifically include:

[0141] Step 1041: Extract the projection geometry mismatch mode information related to the scanning protocol parameters from the constraints, and scan the image region with linear extension morphological features in the original medical image based on the projection geometry mismatch mode information.

[0142] Step 1042: Select target image regions that meet the preset straightness and length thresholds from the image regions with straight-line extension morphological characteristics and mark them as first-class artifact candidate regions.

[0143] Step 1043: Within the first type of artifact candidate region, perform matching filtering based on the artifact morphology template in the projection geometry mismatch mode information to generate a first artifact confidence map.

[0144] Step 1044: Extract tissue density boundary distribution information related to anatomical structure identification information from the constraints, and locate the theoretical boundary regions of different tissue types in the original medical image based on the tissue density boundary distribution information.

[0145] Step 1045: Detect connected regions in the original medical image within the theoretical boundary region that do not conform to the gradient change pattern expected by the tissue density boundary distribution information, and mark the connected regions as second-type artifact candidate regions.

[0146] Step 1046: Within the second type of artifact candidate region, calculate the degree of deviation between the gray value of each pixel in the original medical image and the gray statistical characteristics of the surrounding homogeneous tissue region, and generate a second artifact confidence map.

[0147] Step 1047: Perform a morphological closing operation on the regions in the first artifact confidence map with confidence levels higher than a first preset threshold to connect the broken segments and generate the first artifact artifact map.

[0148] Step 1048: Perform region growing on the regions in the second artifact confidence map where the confidence level is higher than the second preset threshold to fill the holes and generate the second artifact impact map.

[0149] In this step, the first and second types of artifact regions refer to two different types of abnormal image regions identified in the original medical images. The first type is usually related to the scanning equipment parameter settings, while the second type is usually related to the anatomical structural characteristics of the subject being examined. These regions are used for subsequent targeted correction and are identified by artifact detection algorithms.

[0150] Projection geometry mismatch mode information refers to specific artifact mode information derived from scanning protocol parameters that describes the imperfect geometric relationship between the X-ray source, detector, and scanned object. It is used to guide the identification of geometry-related artifacts such as stripe artifacts and is obtained by analyzing the geometric parameters in the scanning protocol.

[0151] An image region refers to a local set of pixels in the original medical image that is analyzed and processed separately. It is used as the basic unit for artifact detection and is obtained through the sliding window technique.

[0152] The preset straightness and length thresholds refer to the pre-set quantitative standards used to determine whether linear structures in an image are artifacts. Straightness measures the straightness of a linear structure, while the length threshold specifies the minimum size of a linear structure. These are used to filter out suspected strip artifacts and are obtained through empirical values ​​or statistical learning.

[0153] The first type of artifact candidate region refers to the image region that is initially screened and may belong to the first type of artifact. It is used for subsequent precise confirmation. It is obtained by applying preset straightness and length thresholds to screen image regions with straight extension morphological features.

[0154] The first artifact confidence map is an image of the same size as the original image. Each pixel value represents the probability that the location belongs to the first type of artifact. It is used to quantify the confidence of the existence of artifacts and is calculated by performing matched filtering within the candidate region of the first type of artifact.

[0155] Tissue density boundary distribution information refers to data obtained from anatomical structure identification information that describes the density transition characteristics between different tissue types. It is used to guide the identification of artifacts caused by abrupt changes in tissue density and is obtained by analyzing standard anatomical atlases.

[0156] The theoretical boundary region refers to the boundary zone between different tissue types as expected based on anatomical knowledge. It is used to locate areas where artifacts such as partial volume effects may occur. It is obtained by mapping the tissue density boundary distribution information onto the image.

[0157] Local grayscale distribution refers to the statistical characteristics of pixel grayscale values ​​within a small region of an image. It is used to analyze whether the image features of that region are abnormal and is obtained by calculating the local region grayscale histogram.

[0158] The second type of artifact candidate region refers to the image region that is initially screened and may belong to the second type of artifact. It is used for subsequent accurate confirmation and is obtained by detecting local gray-level distribution anomalies within the theoretical boundary region.

[0159] The second artifact confidence map is an image of the same size as the original image. Each pixel value represents the probability that the location belongs to the second type of artifact. It is used to quantify the confidence of the existence of artifacts and is obtained by calculating the degree of grayscale deviation within the candidate region of the second type of artifact.

[0160] Grayscale value refers to the brightness value of each pixel in a medical image, used to characterize information such as tissue density, and is obtained directly through image acquisition equipment.

[0161] The first artifact effect map refers to the final determined distribution map of the first type of artifact regions, which is used to guide subsequent artifact suppression operations.

[0162] The second artifact effect map refers to the final determined distribution map of the second type of artifact regions, which is used to guide subsequent artifact suppression operations.

[0163] The first preset threshold refers to the critical value used to extract high-confidence artifact regions from the first artifact confidence map, which is obtained through experience or statistical analysis.

[0164] The second preset threshold refers to the critical value used to extract high-confidence artifact regions from the second artifact confidence map, which is obtained through experience or statistical analysis.

[0165] In this step, the projection geometry mismatch mode information is first obtained by parsing the scanning protocol parameters in the constraints. This information specifically describes the direction and spacing of specific artifacts caused by non-ideal geometric relationships between the X-ray source and the detector, such as misalignment or vibration. Using this information, the image is scanned in the original medical image using a directional filtering algorithm, such as using a set of Gabor filters in a specific direction, to enhance and identify all image regions with linear extension morphological features.

[0166] Secondly, for the identified straight line feature regions, image morphological analysis algorithms such as Hough transform or skeletonization analysis are used to calculate the straightness of each connected region, such as aspect ratio, fitting error with the ideal straight line, and length. The calculation results are compared with preset straightness and length thresholds to filter out regions that meet the conditions of being long and straight enough, and these regions are marked as the first type of artifact candidate regions. Then, within the first type of artifact candidate regions, a template matching filtering algorithm is used to calculate the similarity between the neighborhood of each pixel in the candidate region and the artifact template, such as by using normalized cross-correlation calculation, to generate the first artifact confidence map, where each pixel value represents the probability that the location is a strip artifact.

[0167] Simultaneously, by parsing the anatomical structure identification information in the constraints, tissue density boundary distribution information is obtained, describing the expected density change pattern at the junction of different tissues such as bone and soft tissue. Based on this, the tissue boundary information in the standard anatomical atlas is mapped onto the original medical image using image registration technology to locate the theoretical boundary region. Then, within the theoretical boundary region, local gray-level statistical analysis algorithms, such as calculating the gradient histogram or gray-level distribution variance of the local region, are used to detect whether the local gray-level distribution matches the expected, gently transitioning gradient change pattern based on the tissue density boundary distribution information. Connected regions with abnormal gray-level distribution patterns such as drastic abrupt changes, depressions, or bulges are marked as candidate regions for the second type of artifact.

[0168] Subsequently, within the second type of artifact candidate region, statistical deviation calculation algorithms, such as calculating Mahalanobis distance or Z-score, are used to quantify the degree of deviation of the gray value of each pixel from the gray statistical characteristics of its surrounding homogeneous tissue region, such as the mean and standard deviation. The greater the deviation, the higher the confidence that the point belongs to an artifact, thereby generating a second artifact confidence map. Finally, the two confidence maps are post-processed to optimize the region integrity: for the first artifact confidence map, a morphological closing operation algorithm is used to perform a dilation and erosion operation on a linear structuring element to connect artifact segments that are broken due to noise or incomplete detection, and the regions with confidence higher than the first preset threshold after connection are binarized to generate a complete first artifact response map.

[0169] For the second artifact confidence map, a region growing algorithm is used to grow outwards to regions with similar grayscale characteristics, using pixels with confidence scores higher than a second preset threshold as seed points, to fill the holes in the artifact region and generate a complete second artifact response map.

[0170] For example, following the specific implementation of the previous step, after obtaining the constraints, artifacts in the patient's chest CT image are first identified; secondly, possible projection mismatch patterns are analyzed according to the geometric parameters in the scanning protocol, and several shadows with straight-line characteristics are scanned in the image. Based on preset straightness and length standards, two longer straight shadows are marked as first-type artifact candidate areas; then, template matching is performed within these candidate areas, and the confidence level of each point belonging to a metallic artifact is calculated to generate a first artifact confidence map; simultaneously, based on the chest anatomy atlas, the theoretical boundary areas such as lung tissue and ribs, heart and mediastinum are located, and abnormal gray-level abrupt changes are detected within these areas. Several areas that do not conform to the expected smooth transition are marked as second-type artifact candidate areas; then, the degree of deviation of the gray level of each point within these areas from the surrounding normal tissue is calculated to generate a second artifact confidence map; finally, closure operation and region growing processing are performed on the two confidence maps respectively to obtain a final artifact response map that clearly identifies strip artifacts and partial volume effect artifacts.

[0171] This step effectively distinguishes artifacts from real anatomical structures, reduces false positives, and provides precise localization information for subsequent targeted corrections, thereby improving the accuracy and efficiency of artifact correction and laying a solid foundation for obtaining high-quality medical images.

[0172] Step 105: Suppress the strip artifacts and volumetric effect artifacts of the original medical image according to the artifact effect map to obtain the target medical image, and perform texture consistency repair on the tissue boundary region in the target medical image based on the structural integrity information in the constraint conditions to output the second medical image.

[0173] Optionally, step 105 may specifically include:

[0174] Step 1051: Determine the first correction intensity distribution for the strip artifact region in the original medical image based on the confidence value of each pixel in the first artifact effect image.

[0175] Step 1052: Based on the first correction intensity distribution, construct a strip artifact suppression function, and use the strip artifact suppression function to correct the pixel values ​​of the region in the original medical image corresponding to the first artifact reflection map to generate a preliminary corrected image.

[0176] Step 1053: Based on the area size and confidence value of each connected region in the second artifact effect image, determine the second correction intensity distribution for the volumetric effect artifact region in the preliminary corrected image.

[0177] Step 1054: Based on the second correction intensity distribution, construct a volumetric artifact compensation function, and use the volumetric artifact compensation function to reconstruct pixels in the region corresponding to the second artifact response map in the preliminary corrected image to generate an intermediate corrected image.

[0178] Step 1055: Extract prior information on tissue density distribution from the constraints, perform contrast restoration processing on the tissue contrast distortion areas caused by artifact suppression in the intermediate corrected image based on the prior information on tissue density distribution, and perform global consistency verification to output the target medical image.

[0179] Step 1056: Extract structural integrity information from the constraints. The structural integrity information includes the expected texture direction and grayscale transition rules at the junctions of different tissues.

[0180] Step 1057: Locate the tissue boundary discontinuity region in the target medical image caused by the artifact suppression operation.

[0181] Step 1058: Based on the expected texture direction of the corresponding tissue type in the structural integrity information, construct a texture repair template.

[0182] Step 1059: Collect reference texture samples within the surrounding healthy tissue region of the discontinuous tissue boundary region.

[0183] Step 10510: Determine the priority level of texture repair based on the matching degree between the reference texture sample and the texture repair template.

[0184] Step 10511: Based on the priority level, perform hierarchical texture restoration processing on the discontinuous area at the tissue boundary. For areas in the medical image with a priority level higher than a preset threshold, use a template-guided texture reconstruction method to weightedly fuse the texture restoration template with the reference texture sample to generate a preliminary restored texture. For areas with a priority level lower than the preset threshold, use a sample diffusion texture restoration method to propagate the reference texture sample along the expected texture direction to generate a diffused restored texture.

[0185] Step 10512: Integrate the preliminary repair texture and the diffusion repair texture in the corresponding regions to obtain the integrated tissue boundary region, and apply a smoothing constraint based on the structural integrity information to the integrated tissue boundary region to obtain the repaired region after texture repair.

[0186] Step 10513: Perform local consistency verification on the repaired area to output a second medical image.

[0187] In this step, the artifact suppression map refers to the map generated by the artifact detection algorithm that identifies suspected artifact regions in the image and their confidence levels, and is used to guide subsequent artifact suppression operations.

[0188] Strip artifacts and volumetric artifacts are two common types of image distortion in medical imaging. Strip artifacts usually appear as straight lines of light and dark stripes, while volumetric artifacts appear as blurred tissue boundaries or abnormal grayscale values. They need to be suppressed using specific algorithms and are caused by the physical process of scanning and the characteristics of anatomical structures.

[0189] The target medical image refers to the output image after processing to suppress stripe artifacts and volume effect artifacts, which is used for subsequent texture restoration.

[0190] Structural integrity information refers to prior knowledge describing the texture structure and grayscale transition characteristics that should exist at the junctions of different tissues, which is used to guide texture restoration to maintain anatomical rationality.

[0191] The second medical image refers to the high-quality output image after artifact suppression and texture restoration.

[0192] The confidence value refers to the quantified value of the probability that each pixel in the artifact reflection image belongs to an artifact. It is used to determine the correction strength and is calculated by an artifact detection algorithm.

[0193] The first correction intensity distribution refers to the distribution map of correction intensity to be applied to different regions, determined based on the confidence value of the strip artifacts. It is used to construct the suppression function and is derived by mapping the confidence value.

[0194] The bar artifact suppression function is a mathematical function used to correct bar artifacts, and its behavior is controlled by the correction intensity distribution.

[0195] The preliminary corrected image refers to the intermediate result image after stripe artifact suppression processing, which is used for subsequent processing and is obtained by applying the stripe artifact suppression function.

[0196] The second correction intensity distribution refers to the correction intensity distribution map determined based on the confidence value and region size of the volumetric effect artifact, which is used to construct the compensation function.

[0197] The volumetric effect artifact compensation function is a mathematical function used to correct volumetric effect artifacts. Its behavior is controlled by the correction intensity distribution and is constructed based on the second correction intensity distribution.

[0198] Intermediate corrected images refer to intermediate result images after volumetric artifact compensation processing, used for contrast restoration, and obtained by applying the volumetric artifact compensation function.

[0199] Prior information about tissue density distribution refers to prior knowledge about the normal density range of different tissues, which is used to restore the contrast caused by artifact suppression distortion.

[0200] The expected texture direction refers to the direction of tissue texture expected based on anatomical knowledge. It is used to construct a repair template and is obtained from structural integrity information.

[0201] The grayscale transition law refers to the law of how grayscale should change smoothly at the interface of different tissues. It is used to guide the smoothing constraint and is obtained from structural integrity information.

[0202] Discontinuous tissue boundaries refer to areas of tissue boundary that appear with unnatural textures and breaks in the image after artifact suppression. These areas need to be restored to be continuous and are located in the target medical image using image analysis algorithms.

[0203] A texture restoration template refers to an ideal texture pattern used to guide texture restoration, which is constructed based on the expected texture direction.

[0204] The surrounding healthy tissue region refers to the normal tissue region adjacent to the discontinuous region that is not affected by artifact suppression. It is used as a source of texture samples and is delineated in the target medical image through image segmentation technology.

[0205] Reference texture samples refer to normal texture patterns collected from healthy tissue areas for repair, obtained through sampling operations.

[0206] Texture restoration priority refers to the order in which restoration areas are processed, and is determined based on the matching degree between the reference sample and the restoration template.

[0207] Preliminary texture restoration refers to the texture obtained after restoring high-priority areas using a template-guided method, which is then obtained through weighted fusion of the template and the sample.

[0208] Diffusion-repaired textures refer to textures obtained by repairing low-priority areas using a sample diffusion method, which is achieved by propagating the sample along the texture direction.

[0209] The integrated tissue boundary area refers to the area after merging the two repair textures, which is obtained through texture splicing operations.

[0210] The repaired area after texture restoration refers to the area that has been finally repaired and is ready for output, obtained by applying smoothing constraints.

[0211] In this step, firstly, the confidence value of the first artifact effect image is converted into a first correction intensity distribution through a nonlinear mapping function. Based on this first correction intensity distribution, a strip artifact suppression function is constructed, such as conditional weighted least squares optimization or an adaptive filter. This strip artifact suppression function performs pixel value correction under gradient domain constraints in high-confidence regions and edge-preserving smoothing in low-confidence regions, processing the original image to generate a preliminary corrected image. Secondly, the second artifact effect image is processed through a region attribute analysis algorithm to generate a second correction intensity distribution. Based on this, a volumetric effect artifact compensation function is constructed, such as using an image inpainting-based pixel reconstruction algorithm to perform structure-aware pixel reconstruction on the artifact regions in the preliminary corrected image, generating an intermediate corrected image. Then, through histogram matching or adaptive contrast stretching algorithms, using prior information on tissue density distribution as a target reference, the contrast of the distorted regions in the intermediate corrected image is adjusted, and after global consistency verification, such as detecting block effects or brightness jumps, the target medical image is output.

[0212] Then, the expected texture direction is extracted from the structural integrity information using a direction field estimation algorithm, such as gradient-based principal direction calculation, to construct a texture restoration template. A block matching or sampling algorithm is then used to extract reference texture samples from the surrounding healthy tissue area. Subsequently, a similarity calculation algorithm, such as calculating the structural similarity (SSIM) between the sample and the template, is used to determine the matching degree, and priorities are set. For high-priority areas, a template-guided weighted fusion algorithm, such as linearly mixing the template and sample based on similarity weights, is used to generate a preliminary restored texture. For low-priority areas, a sample diffusion algorithm, such as texture synthesis based on iso-illuminance lines, is used to propagate samples along the expected texture direction to generate a diffused restored texture.

[0213] Then, the preliminary repaired texture and the diffusion repaired texture are integrated by image stitching algorithm. An anisotropic diffusion algorithm is applied to smooth the texture by using grayscale transition rules as constraints to ensure a natural transition between the repaired area and the surrounding area. Finally, the second medical image is output after local consistency verification, such as detecting the consistency of the repaired boundary.

[0214] For example, firstly, after obtaining the artifact response map identifying the strip artifacts and partial volume effect artifacts, the chest CT image of patient B is corrected. Secondly, the correction intensity is determined based on the confidence map of the strip artifacts, and a suppression function is constructed to weaken the two obvious linear artifacts in the image, resulting in a preliminary corrected image. Next, for the partial volume effect artifact area, a compensation function is constructed based on its area and confidence to perform pixel reconstruction on the blurred area at the edge of the cardiac silhouette, making the boundary clearer, resulting in an intermediate corrected image. Subsequently, using prior information on tissue density, the contrast of the mediastinal region, which is slightly distorted due to artifact suppression, is adjusted, and the target medical image is output. Then, chest anatomical structure information is extracted, and it is found that subtle discontinuous textures are generated at the edge of the cardiac silhouette due to correction. Based on this, a texture template is constructed, and texture samples are collected from the surrounding normal myocardial tissue. Then, by calculating the matching degree, the edge of the cardiac silhouette is determined to be a high-priority repair area, and fine repair is performed using a template fusion method. For other low-priority areas, rapid repair is performed using a sample diffusion method. Finally, the repaired textures are integrated and smoothed, and a second medical image with artifact elimination and natural texture is output.

[0215] Step 106: Fuse the first medical image and the second medical image to output the final medical image.

[0216] In this step, the final medical image refers to the high-quality and high-diagnostic-value output image generated by fusing the first medical image, which has been processed by the quality enhancement path and the artifact correction path respectively, with the second medical image, and is used for clinical diagnosis or subsequent analysis.

[0217] In this step, firstly, a feature alignment algorithm is used to spatially register the first and second medical images to ensure that the pixel positions of the same anatomical structure in the two images correspond completely. Secondly, a multi-scale feature extraction algorithm is used to extract feature maps representing different levels of detail from the first and second medical images respectively. Next, an adaptive weight map generation algorithm is used to dynamically calculate the fusion weight of the first or second medical image at each pixel position based on the detail sharpness features of each region in the first medical image and the artifact suppression features of each region in the second medical image, generating a dynamic fusion weight map. Then, a pixel-by-pixel weighted fusion algorithm is used to weight and mix the first and second medical images using the dynamic fusion weight map to generate a preliminary fused image. Subsequently, an edge consistency detection algorithm is used to identify possible stitching marks or unnatural transition areas in the preliminary fused image. Finally, a gradient domain fusion algorithm is used to smooth the detected discontinuous areas to ensure the visual consistency of the entire image, outputting the final medical image.

[0218] For example, following the specific implementation of the previous step, after obtaining the first medical image with enhanced detail and the second medical image with artifact correction and texture restoration, the fusion process begins. Next, the two images are finely registered to ensure complete alignment of key structures such as the heart and lungs. Then, the detail clarity of the lung nodule region in the first medical image and the artifact suppression effect of the mediastinal region in the second medical image are analyzed respectively, and a fusion weight map is generated accordingly—the first medical image is given a higher weight in the nodule region to preserve enhanced details, and the second medical image is given a higher weight in the mediastinal region to maintain artifact suppression. Then, pixel-level weighted fusion is performed, and gradient smoothing is applied to the fusion boundaries, such as the edge of the heart image, to eliminate possible seam traces. Finally, a high-quality chest CT image that clearly shows small nodules and effectively suppresses artifacts is output.

[0219] Figure 2 This application provides a schematic diagram of the structure of an artificial intelligence-based medical image quality enhancement and artifact correction system, as shown below. Figure 2 As shown, the system includes:

[0220] The acquisition module 21 is used to acquire the original medical image of the target object and the constraint conditions matching the original medical image, and to analyze the constraint conditions using a pre-built image processing model to obtain the intensity weight map and continuity guidance signal of the constraint conditions.

[0221] Extraction module 22 is used to perform hierarchical feature extraction and weighted modulation on the original medical image based on the intensity weight map to obtain multi-scale features of the original medical image;

[0222] The construction module 23 is used to construct an intermediate image based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image and the background brightness component of the original medical image, and to perform texture smoothing transition processing on the intermediate image using the continuity guidance signal to output the first medical image.

[0223] The generation module 24 is used to identify the first and second type artifact regions in the original medical image based on the constraints and generate corresponding artifact shadow maps.

[0224] Repair module 25 is used to suppress the strip artifacts and volume effect artifacts of the original medical image according to the artifact effect map, obtain the target medical image, and perform texture consistency repair on the tissue boundary region in the target medical image based on the structural integrity information in the constraint conditions, and output the second medical image.

[0225] The fusion module 26 is used to fuse the first medical image and the second medical image to output the final medical image.

[0226] Figure 2 The aforementioned AI-based medical image quality enhancement and artifact correction system can perform... Figure 1 The implementation principle and technical effects of the AI-based medical image quality enhancement and artifact correction method described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the AI-based medical image quality enhancement and artifact correction system in the above embodiments are performed have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0227] In one possible design, Figure 2 The illustrated embodiment of an AI-based medical image quality enhancement and artifact correction system can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0228] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0229] The processing component 32 is used for the above Figure 1 The embodiment describes an artificial intelligence-based method for enhancing medical image quality and correcting artifacts.

[0230] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for enhancing medical image quality and correcting artifacts based on artificial intelligence, characterized in that, include: The original medical image of the target object and the constraints matching the original medical image are obtained, and the constraints are analyzed using a pre-built image processing model to obtain the intensity weight map and continuity guidance signal of the constraints. Based on the intensity weight map, the original medical image is subjected to hierarchical feature extraction and weighted modulation to obtain multi-scale features of the original medical image; Based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image, and the background brightness component of the original medical image, an intermediate image is constructed. The texture smoothing transition of the intermediate image is then performed using the continuity guidance signal to output a first medical image. The structural consistency enhancement and detail compensation algorithm refers to a set of combined techniques for improving the quality of medical images. This method enhances the clarity of image details while ensuring the natural continuity and integrity of anatomical structures. It is used to generate a high-quality first medical image from the original medical image by coordinating the nonlinear enhancement of detail features and the constrained smoothing of structural features. Identifying first and second type artifact regions in the original medical image based on the constraints and generating corresponding artifact response maps includes: parsing projection geometry mismatch pattern information related to scanning protocol parameters from the constraints; scanning image regions with linear extension morphological features in the original medical image based on the projection geometry mismatch pattern information; selecting target image regions that meet preset straightness and length thresholds from the image regions with linear extension morphological features and marking them as first type artifact candidate regions; performing matching filtering based on artifact morphology templates in the projection geometry mismatch pattern information within the first type artifact candidate regions to generate a first artifact confidence map; parsing tissue density boundary distribution information related to anatomical structure identification information from the constraints and generating corresponding artifact response maps based on the tissue density boundary information. The boundary distribution information is used to locate the theoretical boundary regions of different tissue types in the original medical image; connected regions in the original medical image within the theoretical boundary regions that do not conform to the gradient change pattern expected by the tissue density boundary distribution information are detected, and these connected regions are marked as second-type artifact candidate regions; within the second-type artifact candidate regions, the degree of deviation between the gray value of each pixel in the original medical image and the gray value statistical characteristics of the surrounding homogeneous tissue region is calculated, and a second artifact confidence map is generated; morphological closing operations are performed on regions in the first artifact confidence map with a confidence level higher than a first preset threshold to connect broken segments, generating a first artifact effect map; region growing is performed on regions in the second artifact confidence map with a confidence level higher than a second preset threshold to fill holes, generating a second artifact effect map. Based on the artifact suppression map, the strip artifacts and volume effect artifacts of the original medical image are suppressed to obtain the target medical image. Based on the structural integrity information in the constraints, the texture consistency of the tissue boundary region in the target medical image is restored, and the second medical image is output. The first medical image data and the second medical image data are fused together to output the final medical image.

2. The method according to claim 1, characterized in that, The constraints are analyzed using a pre-built image processing model to obtain the intensity weight map and continuity guidance signal of the constraints, including: The constraints are input into the condition parsing module of the image processing model; The constraint intensity distribution information corresponding to different spatial scales in the constraint conditions is extracted through the first analytical branch in the condition analysis module. The constraint intensity distribution information is scale-aligned and nonlinearly mapped to generate an intensity weight map that matches the pixel position of the original medical image. The topological relationship information characterizing the continuity of the anatomical structure boundary is extracted from the constraint conditions through the second analytical branch in the condition analysis module. Based on the topological relationship information, the expected values ​​of connection strength and transition smoothness between adjacent anatomical regions in the original medical image are calculated. Based on the connection strength and the expected transition smoothness, a continuous guiding signal is generated to guide the smooth transition of local textures in the image.

3. The method according to claim 1, characterized in that, Based on the intensity weight map, hierarchical feature extraction and weighted modulation are performed on the original medical image to obtain multi-scale features of the original medical image, including: The original medical image is input into a preset feature extraction network, and the global contour features, local tissue features and microscopic detail features of the original medical image are extracted by the convolutional layer groups with different receptive fields in the feature extraction network. Separate the first-scale weight distribution, the second-scale weight distribution, and the third-scale weight distribution corresponding to the global contour feature, the local tissue feature, and the microscopic detail feature from the intensity weight map; The first scale weight distribution is multiplied element-wise with the global contour feature to obtain the weighted modulated global contour feature. The second-scale weight distribution is multiplied element-wise with the local tissue features to obtain the weighted modulated local tissue features; The third-scale weight distribution is multiplied element-wise with the micro-detail features to obtain the weighted modulated micro-detail features; The weighted modulated global contour features, the weighted modulated local tissue features, and the weighted modulated microscopic detail features are spliced ​​together according to the feature channel dimension to form the multi-scale features of the original medical image.

4. The method according to claim 1, characterized in that, Based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image, and the background brightness component of the original medical image, an intermediate image is constructed. The intermediate image is then processed using the continuity guidance signal to achieve a smooth texture transition, resulting in the output of a first medical image, including: The weighted modulated microscopic detail features are separated from the multi-scale features as the detail features to be compensated; Based on the consistency of gradient direction of each pixel in the weighted modulated local tissue features, a detail compensation function is constructed. The detail features to be compensated are nonlinearly enhanced by a detail compensation function to generate enhanced detail features. The detailed enhancement features are fused with the weighted modulated local tissue features to obtain the tissue detail fusion features. The structural enhancement features are obtained by concatenating the fused organizational details with the weighted modulated global contour features. The structural enhancement features are superimposed on the background brightness component of the original medical image at the pixel level to generate an initial intermediate image; Extract the transition region in the initial intermediate image that corresponds to the anatomical boundary indicated by the continuous guiding signal; Based on the smoothness expectation value in the continuous guidance signal, the gray values ​​of adjacent pixels in the transition region of the initial intermediate image are subjected to diffusion balance processing to obtain the intermediate image. The intermediate image is subjected to brightness normalization processing to output the first medical image.

5. The method according to claim 1, characterized in that, Based on the artifact suppression map, the stripe artifacts and volume effect artifacts of the original medical image are suppressed to obtain the target medical image, including: Based on the confidence values ​​of each pixel in the first artifact effect image, determine the first correction intensity distribution for the strip artifact region in the original medical image; Based on the first correction intensity distribution, a strip artifact suppression function is constructed. The pixel values ​​of the region corresponding to the first artifact reflection map in the original medical image are corrected by the strip artifact suppression function to generate a preliminary corrected image. Based on the area size and confidence value of each connected region in the second artifact effect diagram, the second correction intensity distribution for the volume effect artifact region in the preliminary corrected image is determined. Based on the second correction intensity distribution, a volumetric artifact compensation function is constructed. The volumetric artifact compensation function is used to reconstruct pixels in the region corresponding to the second artifact correction map in the preliminary corrected image to generate an intermediate corrected image. Prior information on tissue density distribution is extracted from the constraints. Based on the prior information on tissue density distribution, contrast restoration processing is performed on the tissue contrast distortion areas caused by artifact suppression in the intermediate corrected image, and global consistency verification is performed to output the target medical image.

6. The method according to claim 1, characterized in that, Based on the structural integrity information in the constraints, texture consistency restoration is performed on the tissue boundary region in the target medical image, and a second medical image is output, including: Structural integrity information is extracted from the constraints, and the structural integrity information includes the expected texture direction and grayscale transition rules at the junctions of different tissues. Locate the tissue boundary discontinuity region caused by artifact suppression operation in the target medical image; Based on the expected texture direction of the corresponding tissue type in the structural integrity information, a texture repair template is constructed. Reference texture samples were collected within the healthy tissue region surrounding the discontinuous tissue boundary region. The priority level of texture repair is determined based on the matching degree between the reference texture sample and the texture repair template; Based on the priority level, hierarchical texture restoration processing is performed on the discontinuous area at the tissue boundary. For areas in medical images with a priority level higher than a preset threshold, a template-guided texture reconstruction method is used. The texture restoration template and the reference texture sample are weighted and fused to generate a preliminary restored texture. For areas with a priority level lower than the preset threshold, a sample diffusion texture restoration method is used. The reference texture sample is propagated along the expected texture direction to generate a diffused restored texture. The preliminary repair texture and the diffusion repair texture are integrated in the corresponding region to obtain the integrated tissue boundary region. A smoothing constraint based on the structural integrity information is applied to the integrated tissue boundary region to obtain the repaired region after texture repair. Local consistency verification is performed on the repaired area to output a second medical image.

7. An artificial intelligence-based medical image quality enhancement and artifact correction system, applied to the artificial intelligence-based medical image quality enhancement and artifact correction method described in any one of claims 1-6, characterized in that... include: The acquisition module is used to acquire the original medical image of the target object and the constraint conditions matching the original medical image, and to analyze the constraint conditions using a pre-built image processing model to obtain the intensity weight map and continuity guidance signal of the constraint conditions. The extraction module is used to perform hierarchical feature extraction and weighted modulation on the original medical image based on the intensity weight map to obtain the multi-scale features of the original medical image; A construction module is used to construct an intermediate image based on the structural consistency enhancement and detail compensation algorithm, the multi-scale features of the original medical image, and the background brightness component of the original medical image. The intermediate image is then processed by the continuity guidance signal to perform texture smoothing transition processing, and a first medical image is output. The structural consistency enhancement and detail compensation algorithm refers to a set of combined technical methods for improving the quality of medical images. This method enhances the clarity of image details while ensuring the natural continuity and integrity of anatomical structures. It is used to generate a high-quality first medical image from the original medical image by coordinating the nonlinear enhancement of detail features and the constrained smoothing processing of structural features. The generation module is used to identify the first and second type artifact regions in the original medical image based on the constraints and generate corresponding artifact shadow maps. The repair module is used to suppress the strip artifacts and volume effect artifacts of the original medical image according to the artifact effect map, obtain the target medical image, and perform texture consistency repair on the tissue boundary region in the target medical image based on the structural integrity information in the constraint conditions, and output the second medical image. The fusion module is used to fuse the first medical image data and the second medical image data to output the final medical image.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the artificial intelligence-based medical image quality enhancement and artifact correction method as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements an artificial intelligence-based medical image quality enhancement and artifact correction method as described in any one of claims 1 to 6.

Images