Image processing method, computer-aided diagnosis method of tumor, and computing device

Abstract

This application discloses an image processing method, a computer-aided diagnosis method for tumors, and a computing device, relating to large model technology and the field of image processing. The method includes: performing image segmentation on the image to be processed to obtain a mask image of the detection region; determining the spatial location information of the detection region in the image to be processed based on the mask image; performing denoising processing on the image to be processed based on the spatial location information to obtain an initial enhanced image; and performing detail restoration on the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than the display clarity of the detection region in the image to be processed. This application solves the technical problem of poor image processing performance in related technologies.

Description

Technical Field

[0001] This application relates to large model technology and image processing, specifically to an image processing method, a computer-aided diagnosis method for tumors, and a computing device. Background Technology

[0002] Existing image enhancement methods often suffer from problems such as blurred details, edge distortion, or structural distortion when processing low-contrast images. In 3D images, traditional models struggle to balance overall structural stability with clear local details, and are easily affected by noise and registration errors, resulting in poor image processing performance.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides an image processing method, a computer-aided diagnosis method for tumors, and a computing device to at least solve the technical problem of poor image processing performance in related technologies.

[0005] According to one aspect of the embodiments of this application, an image processing method is provided, comprising: performing image segmentation on an image to be processed to obtain a mask image of a detection region, wherein the image to be processed contains a detection region of at least one biological tissue structure; determining spatial positioning information of the detection region in the image to be processed based on the mask image, wherein the spatial positioning information is used to represent the distribution state of the detection region in a corresponding three-dimensional space of the image to be processed; performing denoising processing on the image to be processed based on the spatial positioning information to obtain an initial enhanced image; and performing detail restoration on the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than the display clarity of the detection region in the image to be processed.

[0006] According to one aspect of the embodiments of this application, an image processing method is provided, comprising: responding to an input command applied to an operation interface, displaying an image to be processed on the operation interface, wherein the image to be processed includes a detection region of at least one biological tissue structure; responding to a processing command applied to the operation interface, displaying a target enhancement image on the operation interface, wherein the target enhancement image is obtained by detail restoration of an initial enhancement image based on a preset resolution, the initial enhancement image is obtained by denoising the image to be processed based on spatial positioning information, the spatial positioning information is obtained based on a mask image of the detection region, the mask image is obtained by image segmentation of the image to be processed, the display clarity of the detection region in the target enhancement image is higher than the display clarity of the detection region in the image to be processed, and the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed.

[0007] According to one aspect of the embodiments of this application, an image processing method is provided, comprising: acquiring an image to be processed by calling a first interface, wherein the first interface includes a first parameter, the parameter value of the first parameter including the image to be processed, the image to be processed containing a detection region of at least one biological tissue structure; performing image segmentation on the image to be processed to obtain a mask image of the detection region; determining spatial positioning information of the detection region in the image to be processed based on the mask image, wherein the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; performing denoising processing on the image to be processed based on the spatial positioning information to obtain an initial enhanced image; performing detail restoration on the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than the display clarity of the detection region in the image to be processed; and outputting the target enhanced image by calling a second interface, wherein the second interface includes a second parameter, the parameter value of the second parameter including the target enhanced image.

[0008] According to another aspect of the embodiments of this application, a computing device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0009] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor connected to the memory via a bus for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0010] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0011] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0012] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the methods in various embodiments of this application.

[0013] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0014] In this embodiment, image segmentation is performed on the image to be processed to obtain a mask image of the detection region, wherein the image to be processed contains at least one detection region of biological tissue structure; based on the mask image, the spatial positioning information of the detection region in the image to be processed is determined, wherein the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; based on the spatial positioning information, denoising processing is performed on the image to be processed to obtain an initial enhanced image; detail restoration is performed on the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than the display clarity of the detection region in the image to be processed, thereby improving the image processing efficiency. The purpose of this enhancement is to characterize the spatial distribution of the detection region in the image to be processed by image segmentation. Based on this spatial positioning information, selective denoising is performed on the image to be processed, so that the denoising process only acts on the spatial position corresponding to the detection region, thereby preserving the original characteristics of the non-target region. The initial enhanced image obtained after denoising has higher structural consistency in the detection region. Then, according to the preset resolution, detail restoration is performed on the initial enhanced image, so that the display clarity of the detection region is directionally improved under the constraint of the spatial positioning information. Finally, the goal of achieving higher clarity of the detection region in the target enhanced image than in the image to be processed is achieved, thereby solving the technical problem of poor image processing effect in related technologies.

[0015] It is worth noting that the general description above and the detailed description that follow are merely for illustrative purposes and do not constitute a limitation on this application. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 This is a scene diagram illustrating an image processing procedure according to an embodiment of this application;

[0018] Figure 2 This is a flowchart of an image processing method according to an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of an image processing procedure according to an embodiment of this application;

[0020] Figure 4 This is a flowchart of another image processing method according to an embodiment of this application;

[0021] Figure 5 This is a flowchart of another image processing method according to an embodiment of this application;

[0022] Figure 6 This is a schematic diagram of an image processing apparatus according to an embodiment of this application;

[0023] Figure 7 This is a schematic diagram of another image processing apparatus according to an embodiment of this application;

[0024] Figure 8 This is a schematic diagram of another image processing apparatus according to an embodiment of this application;

[0025] Figure 9 This is a flowchart of a computer-aided diagnosis method for tumors according to an embodiment of this application;

[0026] Figure 10 This is a schematic diagram of another image processing apparatus according to an embodiment of this application;

[0027] Figure 11 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] According to an embodiment of this application, an image processing method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] Considering the large number of model parameters in large models and the limited computing resources of mobile terminals, the method provided in this application embodiment can be applied to, for example, Figure 1 The application scenarios shown are not limited to these. Figure 1 This is a scene diagram illustrating an image processing procedure according to an embodiment of this application, in such a case... Figure 1 In the application scenario shown, the large model is deployed on server 10. Server 10 can connect to one or more client devices 20 via a local area network (LAN), wide area network (WAN), internet connection, or other types of data network. These client devices 20 may include, but are not limited to, smartphones, tablets, laptops, PDAs, personal computers, smart home devices, and in-vehicle devices. Client devices 20 can interact with users through a graphical user interface to access the large model, thereby implementing the method provided in this embodiment.

[0032] In this embodiment, the system consisting of a client device and a server can perform the following steps: The client device performs the step of acquiring an image to be processed. The server performs image segmentation on the image to be processed to obtain a mask image of the detection region; based on the mask image, it determines the spatial positioning information of the detection region in the image to be processed; based on the spatial positioning information, it performs denoising processing on the image to be processed to obtain an initial enhanced image; based on a preset resolution, it performs detail restoration on the initial enhanced image to obtain a target enhanced image.

[0033] It should be noted that with the rapid development of high-performance computing units, the methods provided in this application embodiment can also be applied to model-in-machine systems in other application scenarios. In one optional embodiment, the model-in-machine system has multiple built-in models, and users can select one model to adjust as needed to obtain their own model. The high-performance computing unit built into the model-in-machine system can then directly call the adjusted model to execute the methods provided in this application embodiment. In another optional embodiment, the large model-in-machine system has a pre-trained model built-in, and the high-performance computing unit built into the model-in-machine system can then directly call that model to execute the methods provided in this application embodiment.

[0034] Furthermore, when users need to train their own models, they can upload their own datasets via the client. These datasets are then sent to the server, allowing the server to adjust the pre-trained model using the dataset to obtain the user's customized model, which can then be deployed to the production environment. To facilitate users' model adjustment needs, the server provides complete adjustment tools, development frameworks, and processes, supporting multiple adjustment strategies. This allows the adjusted model to better adapt to different application domains and achieve a high degree of customization.

[0035] Under the aforementioned operating environment, this application provides the following: Figure 2 The image processing method shown. Figure 2 This is a flowchart of an image processing method according to an embodiment of this application. Figure 2 As shown, the method may include the following steps:

[0036] Step S202: Perform image segmentation on the image to be processed to obtain a mask image of the detection area.

[0037] The image to be processed contains a detection area containing at least one biological tissue structure.

[0038] The image to be processed described above can be three-dimensional volumetric data acquired and reconstructed by an imaging device. It consists of multiple voxels, each with unique spatial coordinates and a corresponding intensity value. This intensity value reflects the degree of attenuation of incident radiation by biological tissue. The image has not undergone any contrast enhancement intervention, resulting in small grayscale differences between biological tissue structures, leading to blurred structural boundaries and low local contrast. The acquisition parameters of the image to be processed include slice thickness, interval, scan matrix, and radiation dose. Its data format conforms to the Digital Imaging and Communications in Medicine (DICOM) standard and possesses complete spatial coordinate system and voxel size information.

[0039] The image segmentation described above is a computational process that automatically divides spatial regions based on image content features. By analyzing the intensity distribution, gradient changes, local texture, and spatial connectivity of voxels in the image to be processed, it identifies and separates sub-regions with specific structural features. This process does not rely on prior anatomical knowledge or manual labeling; it outputs the region segmentation results solely based on statistical models and geometric rules. The output is an auxiliary data structure with the same spatial dimensions as the original image.

[0040] The aforementioned detection region is a local three-dimensional spatial subset identified during the image segmentation process that possesses potential structural anomalies or functional significance within the image to be processed. Its extent is defined by a voxel set, and it has a clear spatial boundary and consistent internal density. At least one of the aforementioned biological tissue structures can be the biological tissue structure to be detected and examined.

[0041] The aforementioned mask image is auxiliary data aligned with the spatial dimensions of the image to be processed. The value of each voxel indicates whether the location belongs to the detection region. This mask image does not carry original intensity information; it primarily expresses a binary or probabilistic state. The mask image serves to limit the scope of subsequent processing, ensuring that enhancement operations only apply to designated areas and avoiding irrelevant disturbances to the background biological tissue. The mask image is automatically generated during the image segmentation process, and its data type can be an integer or floating-point three-dimensional array.

[0042] In one optional embodiment, at least one biological tissue structure region with potential structural features can be automatically identified by analyzing the intensity distribution and spatial connectivity of voxels in the image to be processed, and a mask image with the same spatial dimension as the image to be processed can be generated. The mask image identifies the set of voxels in the detection region in a binary or probabilistic form, thereby achieving unsupervised localization of the target region, providing accurate spatial constraints for subsequent processing, avoiding manual intervention, and improving the automation and consistency of the processing flow.

[0043] The images to be processed can be computed tomography (CT) images or non-contrast-enhanced computed tomography (NC-CT) images. In the process of tumor-vascular localization and mask generation, the objective of this application is to automatically identify the key anatomical regions of interest and requiring enhancement, i.e., the detection regions mentioned above, even when the CT contrast is relatively low, thus providing spatial priors for subsequent generation.

[0044] In one optional embodiment, patient NC-CT 3D volumetric data can be input, which may be Digital Imaging and Communications in Medicine (DICOM) or reconstructed volumetric data. The input data is first preprocessed, such as voxel spacing resampling, intensity normalization, and abdominal / pancreatic region cropping. Then, a 3D segmentation network is invoked, employing a neural network structure to automatically segment the pancreas, pancreatic tumors or suspected lesions, and key peripancreatic vessels, outputting a voxel-level mask, i.e., the aforementioned mask image. This mask image can be binary, multi-class, or probabilistic. The mask image may include tumor mask images and vascular mask images.

[0045] The aforementioned key peripancreatic vessels include at least the portal vein and its main branches, the superior mesenteric vein, the superior mesenteric artery, the celiac trunk / common hepatic artery / splenic artery, and the inferior vena cava, which are closely related to resectability assessment.

[0046] When training the image segmentation model used in the image segmentation process, the training mask used during training can be derived from annotations by relevant personnel or an iterative annotation process of "pseudo-labels + expert correction"; during the inference stage, it can be automatically generated by the image segmentation model without human intervention. The tumor mask image and blood vessel mask image output above can be used as display condition inputs for the subsequent diffusion part, to guide the enhancement position, suppress abnormal changes in non-critical regions, and improve the stability of the expression of vascular continuity and the relationship between tumor and blood vessels.

[0047] Step S204: Based on the mask image, determine the spatial location information of the detection area in the image to be processed.

[0048] Among them, spatial positioning information is used to represent the distribution of the detection area in the three-dimensional space corresponding to the image to be processed.

[0049] The aforementioned spatial localization information is a set of mathematical descriptions calculated from the mask image, used to characterize the geometric shape and distribution of the detection region in three-dimensional space. This information includes, but is not limited to, the centroid coordinates of the detection region, the size and orientation of the smallest bounding box, the principal axis direction vector, the boundary curvature distribution statistics, the voxel density gradient field, the minimum distance field with neighboring structures, and the number of connected components. This information does not change the image content; it serves only as an abstract expression of the structural topology, guiding the spatial selectivity and structural consistency of subsequent processing.

[0050] In one optional embodiment, the geometric features of the mask image in the three-dimensional space of the image to be processed can be calculated by analyzing the voxel distribution of the detection area in the mask image, including the centroid position, bounding box size, principal axis direction and boundary curvature, etc., to form spatial positioning information describing the spatial shape and distribution state of the region, providing structural context for subsequent processing, ensuring that denoising and enhancement operations can respond directionally according to the real spatial characteristics of the region, and improving the local accuracy and consistency of the processing.

[0051] Step S206: Denoise the image to be processed based on spatial positioning information to obtain an initial enhanced image.

[0052] The denoising process described above is a computational process that adjusts the intensity locally based on spatial positioning information. Through filtering or generative models, it can reduce the interference of random noise in the image being processed while preserving structural boundaries and key texture features. This denoising process dynamically adjusts the filtering intensity and range based on spatial positioning information, performing intensity correction in the detection area and its adjacent boundary regions, thereby avoiding structural blurring or distortion in the background area. The image output by this process exhibits higher contrast stability and lower noise levels within the detection area.

[0053] The initial enhanced image described above is 3D volumetric data generated after denoising. Its voxel resolution remains consistent with the image to be processed, but the intensity distribution within the detection area has been locally adjusted based on spatial positioning information, achieving initial contrast enhancement. This image does not introduce new structural morphology; it only improves the grayscale response of existing structures, making their edges clearer and their internal density distribution more consistent, providing a structural basis for subsequent detail restoration. This image still lacks high-frequency texture and edge sharpness; its enhancement effect belongs to the low-frequency level structural consistency improvement.

[0054] In one alternative embodiment, the denoising process can be guided to perform intensity adjustment within the detection area and its adjacent boundaries based on the three-dimensional distribution state of the detection area in the image to be processed as described by the spatial positioning information. This suppresses noise interference while preserving the integrity of the structural edges, making the gray-scale distribution within the detection area more stable and outputting an initial enhanced image with higher contrast consistency, thereby determining the structural basis for subsequent detail restoration.

[0055] Step S208: Perform detail restoration on the initial enhanced image based on the preset resolution to obtain the target enhanced image.

[0056] Among them, the display clarity of the detection area in the target enhancement image is higher than that of the detection area in the image to be processed.

[0057] The aforementioned preset resolution is the final spatial sampling density required for the target enhanced image. Its value is preset by the imaging system output specifications or downstream display and interpretation requirements, and is usually higher than the current voxel density of the initial enhanced image. This parameter is a system-level configuration parameter and does not depend on the original acquisition resolution. Its function is to ensure that the generated image meets the sharpness standards required for visual interpretation. The preset resolution is achieved through interpolation or reconstruction algorithms, and its unit is voxels / mm, corresponding to the sampling intervals in the axial, coronal, and sagittal directions.

[0058] The aforementioned detail restoration is a high-frequency information reconstruction process based on a nonlinear mapping model. It restores edge sharpness, texture grains, microstructural contours, and the natural distribution patterns of acquisition noise lost due to noise suppression, downsampling, or low-frequency filtering by enhancing the resolution and texture of the initial enhanced image. This process does not introduce new anatomical structures; it only supplements visual details based on existing low-frequency structures. Adversarial generative networks or super-resolution reconstruction models can be used to make the output image more perceptually close to the true enhanced state.

[0059] The aforementioned enhanced image is the final 3D volumetric data generated during the detail restoration process. Its spatial resolution reaches the preset resolution, the boundaries of the detection area are clear, the texture is rich, the contrast distribution is reasonable, and the overall structure remains consistent with the original anatomical morphology. This image can be effectively recognized by human readers, its clarity is higher than the image to be processed, and no artifacts, structural distortions, or non-physiological enhancements are observed.

[0060] In one optional embodiment, spatial interpolation and high-frequency texture reconstruction can be performed on the initial enhanced image based on a preset resolution to restore the edge sharpness and fine structure texture weakened by denoising. This allows the detection area to present a clearer visual appearance while maintaining its original shape, thereby improving the recognizability of the target enhanced image at the target resolution without introducing false structures.

[0061] Image segmentation can be performed on the image to be processed to obtain a mask image containing at least one biological tissue structure for detecting the target area, thus clarifying the spatial extent of the target area. Based on the mask image, the distribution of the detection area in the three-dimensional space of the image to be processed is calculated to form spatial positioning information, providing a structural basis for subsequent processing. Subsequently, local denoising is performed on the original image based on the spatial positioning information to suppress noise interference and preserve morphological features, resulting in a more structurally stable initial enhanced image. Finally, high-frequency detail reconstruction is performed on the initial enhanced image at a preset resolution, making the boundaries and textures of the detection area clearer. This achieves higher display clarity of the detection area in the target enhanced image than in the original image to be processed, improving structural visibility without introducing non-realistic content.

[0062] For example, the abdominal scan image to be processed is segmented, and the pancreatic region is identified as the detection area to generate a corresponding mask image. Based on the mask image, the position, shape and boundary range of the pancreas in three-dimensional space are determined to form spatial positioning information. Subsequently, denoising is performed only on the pancreas and its adjacent areas, preserving its outline without blurring key structures such as blood vessels, to obtain an initial enhanced image. Finally, the initial image is reconstructed at the required resolution to make the biological tissue texture and edges inside the pancreas clearer, so that the display clarity of the pancreas in the target enhanced image is significantly higher than that of the original image to be processed, which helps to observe subtle changes.

[0063] The goal of the above steps is to learn the enhancement rules of the image while maintaining the consistency of the anatomical structure, and to generate a contrast representation consistent with the actual arterial / venous phase. To balance memory usage and efficiency, the input volume data can be downsampled from high resolution to medium resolution (e.g., from 512×512 to 256×256) in the axial plane, and 3D generation calculations can be performed at this resolution.

[0064] The above process can employ a conditional diffusion framework, which can be triggered by random noise during generation and gradually "de-noise" and "refine" through multiple iterations until the virtual augmentation data of the target phase is obtained, which is the initial augmented image mentioned above.

[0065] The conditional information used in the aforementioned conditional diffusion framework can consist of two parts. The first part can be the original NC-CT volume data, i.e., the image to be processed, which provides the basis for the patient's individual anatomy and density distribution. The second part can be the tumor and peripancreatic vascular mask, i.e., the spatial localization information provided by the mask image, which provides spatial priors. To ensure that the mask image truly participates in the generation control rather than simply being stitched together, a spatial adaptive modulation mechanism can be introduced in the network normalization and feature modulation stages. The mask image can be encoded into spatial guidance parameters at the same scale as the feature map, i.e., the aforementioned spatial localization information. By differentially adjusting the features at different spatial locations, the conditional diffusion framework can learn enhancement patterns that are more consistent with clinical patterns in the vascular lumen, vascular wall, pancreatic parenchyma, and tumor periphery.

[0066] This application can be used for intelligent enhancement of non-contrast-enhanced CT (NC-CT) images. First, pancreatic tumors and peripancreatic vessels are segmented using a segmentation model to generate a mask for precise localization of the spatial distribution of lesions and vessels. Then, noise reduction is guided by the mask to preserve key anatomical structures while suppressing noise, resulting in an initial enhanced image. Finally, a super-resolution generative adversarial network (GAN) is used to restore high-resolution details, improving the clarity of tumor edges and vessel contours, and approximating the effect of real contrast-enhanced CT. This provides reliable support for improving diagnostic confidence in pancreatic cancer detection and resectability assessment in contrast-free scenarios.

[0067] Through the above steps, image segmentation is performed on the image to be processed to obtain a mask image of the detection region, wherein the image to be processed contains at least one detection region of biological tissue structure; based on the mask image, the spatial positioning information of the detection region in the image to be processed is determined, wherein the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; based on the spatial positioning information, denoising processing is performed on the image to be processed to obtain an initial enhanced image; detail restoration is performed on the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than that of the detection region in the image to be processed, thereby improving image processing efficiency. The purpose of this effect is to characterize the spatial distribution of the detection region in the image to be processed by image segmentation. Based on this spatial positioning information, selective denoising is performed on the image to be processed, so that the denoising process only acts on the spatial position corresponding to the detection region, thereby preserving the original characteristics of the non-target region. The initial enhanced image obtained after denoising has higher structural consistency in the detection region. Then, according to the preset resolution, detail restoration is performed on the initial enhanced image, so that the display clarity of the detection region is directionally improved under the constraint of spatial positioning information. Finally, the goal is to achieve the purpose that the clarity of the detection region in the target enhanced image is higher than that in the image to be processed, thereby solving the technical problem of poor image processing effect in related technologies.

[0068] In the above embodiments of this application, the process of restoring details of an initial enhanced image based on a preset resolution to obtain a target enhanced image includes: enlarging the initial enhanced image to a preset resolution to obtain an enlarged image; and using an image enhancement model to restore details of the enlarged image to obtain the target enhanced image.

[0069] The magnified image mentioned above is intermediate data obtained by spatially expanding the initial enhanced image to a preset resolution using an interpolation method.

[0070] The image enhancement model described above is a nonlinear mapping network based on deep learning. Its training objective is to recover high-frequency structural details from low-resolution images. Here, it is used to enlarge the image to reconstruct edges and textures that have been blurred due to interpolation.

[0071] In one optional embodiment, the initial enhanced image can be spatially interpolated at a preset resolution to obtain a magnified image. This process expands the voxel density of the image to match the target scale, but it is accompanied by edge blurring and texture loss. Subsequently, the magnified image is nonlinearly reconstructed using an image enhancement model. This image enhancement model is based on learned structural patterns and performs high-frequency detail filling in the blurred areas introduced by interpolation, restoring the boundary sharpness and subtle textures weakened by the size expansion, thereby generating a target enhanced image, making the structural representation of the detection area clearer and improving visual recognizability.

[0072] The initial enhanced image is magnified to a preset resolution to obtain a magnified image. This process increases the number of pixels through interpolation to make the image size match the target scale, but only extends the spatial sampling and does not restore the original details. Subsequently, the magnified image is restored using an image enhancement model. Based on the learned local structural rules, the image enhancement model reconstructs high-frequency textures and boundary transitions without changing the overall shape, so that the magnified distorted local features tend to be clear in the original state. The final target enhanced image maintains structural consistency while improving the distinguishability of regional boundaries and enhancing the stability of subsequent processing or observation.

[0073] In the process of performing axial slice super-resolution and detail enhancement, this application can further supplement high-frequency details on the initial enhanced image without destroying the three-dimensional anatomical consistency already established in the initial enhanced image. This makes the final generated target enhanced image closer to the real image in terms of edge sharpness, vascular contour continuity, texture and noise appearance, thereby improving the clinical image reading experience and readability.

[0074] In the above embodiments of this application, the denoising process of the image to be processed based on spatial positioning information to obtain an initial enhanced image includes: controlling a denoising model based on spatial positioning information to denoise the image to be processed to obtain an initial enhanced image.

[0075] In one optional embodiment, the distribution of the mask image in three-dimensional space can be calculated based on the detection region boundary identified by the mask image to form spatial positioning information. This spatial positioning information defines the area range and intensity distribution of the denoising effect. Secondly, the spatial positioning information is input as a control parameter into the denoising model, so that the denoising model dynamically adjusts the filtering intensity, radius of action, or weight distribution based on the spatial positioning information when performing denoising operations. This allows for targeted denoising in the detection region and its adjacent regions, avoiding structural blurring or loss of detail in non-target regions. Finally, the spatially constrained denoising result is output, i.e., the initial enhanced image.

[0076] This application embodiment uses spatial positioning information to perform regional adaptive control of the denoising model, thereby achieving accurate noise reduction of the detection area. It effectively avoids the structural damage caused to non-target areas by traditional global denoising methods, and maintains the boundary clarity and morphological consistency between the detection area and the surrounding structure while reducing image noise.

[0077] In the above embodiments of this application, the denoising model based on spatial positioning information is used to denoise the image to be processed to obtain an initial enhanced image, including: extracting features from the image to be processed to obtain image texture features; performing feature processing on spatial positioning information based on the feature scale of the image texture features to obtain spatial guidance features; and denoising the image to be processed based on the image texture features and spatial guidance features to generate an initial enhanced image.

[0078] The aforementioned image texture features are local spatial variation patterns extracted from the image to be processed. They are used to reflect the gray-scale distribution patterns and structural repeatability of different regions, and to characterize the intrinsic uniformity, boundary sharpness, and detail richness of biological tissues or materials.

[0079] The aforementioned feature scale represents the level of variation of image texture features in different spatial ranges. The feature scale is used to describe the local complexity and response range of the texture structure and to determine the adaptive granularity of subsequent processing.

[0080] The aforementioned spatial positioning information is a geometric representation calculated based on the mask image, describing the location, shape, and distribution of the detection area in three-dimensional space, and providing structural guidance for the processing.

[0081] The aforementioned spatial guidance feature is a dynamic weight distribution generated by matching and modulating the feature scale of spatial positioning information, which coordinates with the spatial variation law of image texture features and is used to control the region selection and intensity distribution of the denoising operation.

[0082] In one optional embodiment, firstly, image texture features are extracted from the image to be processed to quantify the structural change characteristics of each region; secondly, based on the feature scale represented by the image texture features, spatial positioning information is adaptively modulated to generate spatial guidance features that match the texture changes; finally, the image texture features and spatial guidance features are used together as control criteria to guide the denoising process to take differentiated action in different regions, thereby achieving synergistic improvement of structure preservation and noise suppression, and generating an initial enhanced image.

[0083] By extracting features from the image to be processed, image texture features are obtained. This process uses multi-scale convolution operations to capture local gray-level changes and structural patterns, reflecting the fine distribution characteristics inside biological tissues. Based on the feature scale of the image texture features, spatial positioning information is processed, aligning the detection region boundary defined by the mask with the texture scale to generate spatially guided features, so that the denoising process focuses on structural edges rather than uniform regions. Based on the image texture features and spatially guided features, the image to be processed is denoised, suppressing non-structural noise while preserving the real biological tissue structure, generating an initial enhanced image. This image maintains the original morphological integrity of the detection region while reducing interference, providing a more reliable input basis for subsequent high-resolution reconstruction.

[0084] This application embodiment forms spatial guidance features through the coordinated modulation of image texture features and spatial positioning information. This enables the denoising process to dynamically adjust its action strategy based on the complexity of the local structure, effectively maintaining the morphological integrity of the detection area while suppressing noise, and improving the structural fidelity and visual consistency of the processing results.

[0085] In one optional embodiment, to improve the stability of large-volume CT training and generation, a dynamic dual-output approach can be adopted. The network not only predicts the denoising information required for traditional diffusion but also predicts content information closer to the target enhanced image itself in parallel. The contributions of both are adaptively balanced through learnable fusion weights, thus simultaneously ensuring structural stability and texture realism in different cases and regions. Structural stability means avoiding distortion and anatomical alterations, while texture realism means avoiding blurring and ensuring clear boundaries. To address the issue of multi-phase CT registration failures in the abdominal cavity and intestinal gas regions, self-regulatory constraints can be added to suppress areas prone to registration inconsistencies, reducing the risk of the model learning registration errors as changes requiring enhancement. The final output consists of medium-resolution arterial and venous phase virtual enhancer data, i.e., the initial enhanced images mentioned above, serving as the basis for subsequent detail restoration.

[0086] In the above embodiments of this application, the denoising process of the image to be processed based on image texture features and spatial guidance features to generate an initial enhanced image includes: iteratively denoising the image to be processed based on spatial guidance features to obtain a denoised image; performing texture enhancement on the denoised image based on image texture features to obtain a texture enhanced image; and generating an initial enhanced image based on the denoised image and the texture enhanced image.

[0087] The iterative denoising described above is a processing method that gradually reduces unstructured interference components in an image by repeatedly applying denoising operators. Each iteration uses spatially guided features as constraints to dynamically adjust the denoising intensity and range, so that the noise elimination process can be gradually advanced while preserving structural continuity.

[0088] The aforementioned image texture features are local grayscale variation patterns extracted from the image to be processed. They are used to characterize the structural repeatability, boundary sharpness, and detail richness of different regions. Their scale and distribution reflect the inherent structural complexity of the image.

[0089] The texture enhancement described above is a local contrast enhancement operation based on image texture features. By enhancing the grayscale gradient changes, it makes the subtle structural features weakened by denoising reappear without introducing new structures or changing the original form.

[0090] The texture enhancement image mentioned above is the output of the texture enhancement operation. The texture enhancement image improves the detail contrast of local areas and enhances the edge sharpness, but does not produce false structures or over-sharpening artifacts, serving as auxiliary information for structure preservation and detail enhancement.

[0091] In one optional embodiment, spatial constraints can be applied to the denoising model based on spatial guidance features, and multiple iterations of denoising can be performed on the image to be processed to gradually generate a denoised image with reduced noise and preserved structure. Next, the image texture features of the denoised image are extracted, and texture enhancement operations are performed on its local change patterns to generate a texture-enhanced image with enhanced detail visibility. Finally, the denoised image and the texture-enhanced image are weighted and fused to combine their advantages in noise suppression and detail restoration to generate an initial enhanced image.

[0092] Iterative denoising of the image under processing is performed based on spatially guided features. By successively updating pixel values ​​and using spatial positioning information as constraints, random fluctuations in unstructured regions are gradually suppressed while preserving boundary continuity, resulting in a denoised image. Subsequently, texture enhancement is performed on the denoised image based on image texture features. Local grayscale variation patterns are used to enhance the contrast of fine structures, making the texture distribution clearer without introducing artifacts, resulting in a texture-enhanced image. Finally, the denoised image and the texture-enhanced image are fused according to weights, taking into account both structural integrity and detail representation, to generate an initial enhanced image. This image reduces interference while maintaining the distribution pattern of the original features, providing a more stable and information-rich input foundation for subsequent high-resolution reconstruction.

[0093] This application embodiment achieves a balance between noise suppression and detail restoration by using spatially guided iterative denoising and texture feature-based enhancement operations in a step-by-step coordinated manner. This allows the denoising process to prioritize the protection of structural boundaries, while the texture enhancement process to specifically restore subtle features. The final fusion result effectively improves local image contrast and structural discernibility while reducing noise.

[0094] In the above embodiments of this application, the method further includes: acquiring a first sample image to be processed, sample spatial positioning information of the first sample image to be processed, and a sample enhancement image corresponding to the first sample image to be processed; controlling an initial denoising model to denoise the first sample image to be processed based on the sample spatial positioning information to obtain a predicted enhancement image; constructing a diffusion loss function based on the sample enhancement image and the predicted enhancement image; and adjusting the model parameters of the initial denoising model based on the diffusion loss function to obtain a denoising model.

[0095] The first sample image to be processed is the original three-dimensional volume data from the imaging device. Its voxel value reflects the original measurement results of physical quantities and contains mixed information of structure, noise and low contrast features, which serves as the input sample for model training.

[0096] The aforementioned sample spatial positioning information is a geometric representation aligned with the spatial representation of the first sample image to be processed. It is calculated from the mask image and is used to identify the position, shape, and distribution of the detection area in three-dimensional space, providing structural guidance for the model.

[0097] The aforementioned sample augmentation image is a high-quality target image paired with the first sample image to be processed. It has a low noise level and clear structural boundaries, representing an ideal processing result, and is used to supervise the model's learning of the mapping relationship from low-quality input to high-quality output.

[0098] The aforementioned initial denoising model is a parameterized image processing network capable of converting the image to be processed to the target quality. Its structure has not been trained and adjusted, and its initial parameters are random or preset values. It has not yet learned the denoising rules of a specific image domain.

[0099] The predicted enhanced image mentioned above is the output result of the initial denoising model after performing denoising processing on the first sample image to be processed. Its quality is affected by the model parameters and has not yet been aligned with the sample enhanced image. It serves as an intermediate output for loss calculation.

[0100] The aforementioned diffusion loss function is an improved objective function based on image differences. By measuring the degree of deviation between the predicted enhanced image and the sample enhanced image in terms of overall structure and local texture, it guides the model parameters to approximate the true distribution. Its design does not rely on pixel-level point-by-point matching, but focuses on global distribution consistency.

[0101] The denoising model described above is the initial denoising model after training and adjustment. Its parameters have been improved by minimizing the diffusion loss function, and it can generate output results that preserve structure and suppress noise when new samples are input.

[0102] In one optional embodiment, a first sample image to be processed, the sample spatial location information corresponding to the first sample image to be processed, and a paired sample augmentation image can be obtained to construct a training data pair. Next, using the sample spatial location information as a condition, an initial denoising model is driven to perform denoising processing on the first sample image to be processed to generate a predicted augmentation image. Then, the predicted augmentation image and the sample augmentation image are compared to construct a diffusion loss function to quantify the differences between the two in terms of structure and texture distribution. Finally, based on the gradient information of the diffusion loss function, the internal parameters of the initial denoising model are iteratively adjusted until the model output approaches the sample augmentation image, forming a denoising model with generalization ability.

[0103] The sample spatial localization information is generated from the mask image of the detection area and is used to limit the denoising range. Based on the sample spatial localization information, the initial denoising model is controlled to denoise the first sample image to be processed, so that the model only performs noise suppression within the area defined by the mask, avoiding interference with the background area, thereby generating a predicted enhanced image. Based on the pixel-level difference between the sample enhanced image and the predicted enhanced image, a diffusion loss function is constructed. This function guides the model to gradually approach the real enhancement effect by calculating the deviation of the image structure and texture layer by layer. Based on the diffusion loss function, the model parameters of the initial denoising model are iteratively adjusted so that the model can more accurately restore the real denoised form while preserving the structural continuity of the detection area. Finally, a denoising model is obtained. In subsequent processing, this denoising model can selectively eliminate noise based on the sample spatial localization information, improving the structural fidelity and consistency of the initial enhanced image.

[0104] This application's embodiments introduce sample spatial positioning information as a condition constraint and combine it with the diffusion loss function to guide the model to learn the distribution-level mapping relationship. This enables the denoising model to achieve semantic enhancement of structural regions without requiring pixel-level precise alignment, thereby improving the coordination and consistency of the output image in terms of noise suppression and boundary preservation.

[0105] In the above embodiments of this application, the first sample image to be processed is denoised based on the sample spatial positioning information to obtain a predicted enhanced image. This includes: extracting features from the first sample image to be processed to obtain predicted image texture features; performing feature processing on the sample spatial positioning information based on the feature scale of the predicted image texture features to obtain predicted spatial guidance features; iteratively denoising the first sample image to be processed based on the predicted spatial guidance features to obtain a predicted denoised image; enhancing the texture of the predicted denoised image based on the predicted image texture features to obtain a predicted texture enhanced image; and generating a predicted enhanced image based on the predicted denoised image and the predicted texture enhanced image.

[0106] The aforementioned predicted image texture features are local grayscale variation patterns extracted from the first sample image to be processed. They are used to characterize the structural repeatability, boundary sharpness, and detail richness of different regions in the image. Their scale and distribution reflect the inherent structural complexity and have not undergone any enhancement or filtering processing.

[0107] The aforementioned feature scale represents the level of change in the texture features of the predicted image across different spatial ranges. It describes the local response range and granularity of the texture structure and determines the modulation granularity of subsequent spatially guided features.

[0108] The aforementioned predictive spatial guidance feature is a dynamic spatial weight distribution generated by adaptively modulating the spatial positioning information of the sample based on the feature scale of the predicted image texture features. Its value changes with the local texture complexity, enhancing the protection weight in the structural boundary region and reducing the constraint intensity in the uniform region, thereby achieving spatial selectivity in the denoising effect.

[0109] The predicted denoised image mentioned above is the output of the iterative denoising process. Its overall noise level is lower than that of the first sample image to be processed. The structural edges and biological tissue morphology are not significantly blurred under the guidance of the predicted spatial features, providing a structurally stable basic input for subsequent texture restoration.

[0110] The texture enhancement described above is a local contrast enhancement operation based on predicted image texture features. By enhancing the gray-level gradient changes, it makes the subtle structural features weakened by denoising reappear without introducing new structures or changing the original form.

[0111] The predicted texture enhancement image mentioned above is the output of the texture enhancement operation. Its local area detail contrast is improved and edge sharpness is enhanced, but no false structure or over-sharpening artifacts are produced. It serves as auxiliary information for structure preservation and detail enhancement.

[0112] The aforementioned predicted enhanced image is an intermediate output generated by weighted fusion of the predicted denoised image and the predicted texture enhanced image. It restores the texture details weakened by smoothing while suppressing noise, and the overall structural integrity and recognizability are synergistically improved.

[0113] In one optional embodiment, predicted image texture features can be extracted from a first sample image to be processed to quantify its local structural change patterns; based on the feature scale represented by the predicted image texture features, the sample spatial positioning information is modulated to generate predicted spatial guiding features that match the local texture complexity of the image; with the predicted spatial guiding features as constraints, multiple iterations of denoising are performed on the image to be processed to generate a predicted denoised image; based on the predicted image texture features, a texture enhancement operation is performed on the predicted denoised image to generate a predicted texture enhanced image; the predicted denoised image and the predicted texture enhanced image are fused to generate a predicted enhanced image.

[0114] Feature extraction can be performed on the first sample image to be processed to obtain the texture features of the predicted image. This process captures local gray-level changes and structural distribution patterns through multi-scale convolution operations. Based on the feature scale of the texture features of the predicted image, feature processing is performed on the spatial positioning information of the sample to adjust the response intensity and spatial distribution of the mask region, resulting in predicted spatial guiding features, which align the denoising range with the texture region. Based on the predicted spatial guiding features, the first sample image to be processed is iteratively denoised, gradually eliminating unstructured noise while preserving the overall structural morphology, generating a predicted denoised image. Based on the texture features of the predicted image, the predicted denoised image is texture-enhanced, strengthening local gray-level transitions and edge sharpness, making fine structures clearer without introducing artifacts, generating a predicted texture-enhanced image. Based on the predicted denoised image and the predicted texture-enhanced image, a weighted fusion is performed to balance structural integrity and detail representation, generating a predicted enhanced image. This image maintains the distribution pattern of the original features while suppressing interference, providing a more stable input basis for subsequent high-resolution reconstruction.

[0115] This application embodiment generates predictive spatial guidance features by co-modulating the predicted image texture features and sample spatial positioning information. This guides the iterative denoising and texture enhancement process to achieve adaptive response in structurally sensitive areas, enabling the predicted enhanced image to achieve a dynamic balance between noise suppression and detail restoration, thereby improving the model's ability to perceive low-contrast areas.

[0116] In the above embodiments of this application, adjusting the model parameters of the initial denoising model based on the diffusion loss function to obtain a denoising model includes: obtaining a sample denoised image and a sample texture enhanced image corresponding to the first sample image to be processed; constructing a denoising loss function based on the sample denoised image and the predicted denoised image; constructing a texture enhancement loss function based on the sample texture enhanced image and the predicted texture enhanced image; and adjusting the model parameters of the initial denoising model based on the diffusion loss function, the denoising loss function, and the texture enhancement loss function to obtain a denoising model.

[0117] The above-mentioned sample denoised images are high-quality intermediate outputs paired with the first sample image to be processed. Their noise level is lower than that of the original input, the structural boundaries are clear and not blurred, and they represent the reference result of ideal denoising processing.

[0118] The aforementioned sample texture enhancement image is a local contrast enhancement version corresponding to the sample denoising image. While preserving structural integrity, it enhances grayscale gradient changes to reveal subtle texture features weakened by denoising, without introducing false structures or artifacts.

[0119] The predicted denoised image mentioned above is an intermediate output generated by the initial denoising model after iterative denoising of the first sample image to be processed. Its quality is affected by the model parameters and has not yet been aligned with the sample denoised image. It is used as the evaluation object for denoising loss calculation.

[0120] The aforementioned predicted texture-enhanced image is an auxiliary output generated after performing texture enhancement operations based on the predicted denoised image and image texture features. Its local detail contrast is improved, but the structure is not deformed. It is used as the evaluation object for calculating texture enhancement loss.

[0121] The aforementioned denoising loss function is an improved objective function that measures the structural differences between the predicted denoised image and the sample denoised image. Its calculation is based on the deviation of the image in the overall gray-level distribution, boundary continuity and noise residue, and is used to guide the model to improve its denoising ability.

[0122] The texture enhancement loss function described above is an improved objective function that measures the difference in local texture consistency between the predicted texture enhancement image and the sample texture enhancement image. It focuses on edge sharpness, texture contrast and detail recovery, and is used to guide the model to effectively recover identifiable structural features after denoising.

[0123] The aforementioned diffusion loss function is an improved objective function based on the overall image distribution consistency. By comparing the similarity between the predicted enhanced image and the sample enhanced image in multi-scale structure and texture distribution, it avoids pixel-level matching bias and guides the model to learn higher-order semantic mapping relationships.

[0124] The internal parameters of the denoising model have been stably converged by jointly improving the diffusion loss function, the denoising loss function, and the texture enhancement loss function, and it can generate output results with clear structure, noise suppression, and coordinated texture enhancement when new samples are input.

[0125] In one optional embodiment, a sample denoised image and a sample texture-enhanced image corresponding to a first sample image to be processed can be obtained as a reference benchmark for the supervision signal. Next, the first sample image to be processed is subjected to denoising and texture enhancement processing using an initial denoising model to generate a predicted denoised image and a predicted texture-enhanced image. Then, a denoising loss function and a texture enhancement loss function are constructed respectively to quantify the degree of structural and texture deviation between the predicted result and the sample reference. Finally, the diffusion loss function, the denoising loss function, and the texture enhancement loss function are combined to drive the parameter update of the initial denoising model, enabling the model to improve simultaneously in terms of global distribution consistency, local structure fidelity, and detail recovery capability, forming the final denoising model.

[0126] The system can obtain the sample denoised image and sample texture enhanced image corresponding to the first sample image to be processed, as a real reference. Based on the pixel-level difference between the sample denoised image and the predicted denoised image, a denoising loss function is constructed to constrain the model to accurately restore the denoising result while preserving structural integrity. Based on the local gray-level transition and edge consistency error between the sample texture enhanced image and the predicted texture enhanced image, a texture enhancement loss function is constructed to guide the model to avoid introducing false structures or blur artifacts when enhancing details. By combining the diffusion loss function, the denoising loss function, and the texture enhancement loss function, the model parameters of the initial denoising model are jointly improved, so that the model can simultaneously coordinate the three objectives of noise suppression, texture restoration, and overall structure preservation during the iteration process, and finally obtain a denoising model. This model can adaptively generate prediction results that are more consistent with the sample denoised image and sample texture enhanced image based on the inherent features of the input image, thereby improving the structural realism and visual continuity of the enhanced output.

[0127] This application embodiment improves the denoising loss function, texture enhancement loss function and diffusion loss function by jointly improving the model in three dimensions: structural fidelity, detail restoration and global distribution consistency. This enables the denoising model to generate enhancement results with low noise, high contrast and natural texture even under unsupervised paired image enhancement, thus improving the ability to recognize low contrast regions.

[0128] In the above embodiments of this application, the method further includes: obtaining a second sample image to be processed and a sample development image corresponding to the second sample image to be processed, wherein the sample development image is an image generated after injecting a target medium into at least one biological tissue structure; predicting the second sample image to be processed using an initial enhancement model to obtain a predicted enhancement image; constructing an image loss function based on the sample development image and the predicted enhancement image; and adjusting the model parameters of the initial enhancement model based on the image loss function to obtain an image enhancement model.

[0129] The second sample image to be processed mentioned above is the original three-dimensional volume data from the imaging device. Its voxel value reflects the physical response characteristics of biological tissue in the absence of a target medium. It contains a mixture of structural information and low soft biological tissue contrast features, and serves as input data under contrast agent-free conditions.

[0130] The aforementioned sample imaging images are enhanced three-dimensional volumetric data acquired by an imaging device after the target medium is injected into at least one biological tissue structure. The changes in voxel values ​​are due to the changes in physical properties caused by the spatial distribution of the medium within the biological tissue.

[0131] The aforementioned predicted enhanced image is a simulated enhanced image output by the initial enhancement model after performing image transformation on the second sample image to be processed. Its grayscale distribution and structural morphology simulate the enhancement effect after the target medium is injected. It has not yet been aligned with the sample development image and serves as an intermediate output for loss calculation.

[0132] The image loss function described above is an improved objective function that measures the degree of deviation between the predicted enhanced image and the sample developed image in terms of overall structure, contrast distribution, and texture consistency. Its design is based on pixel-level or high-order feature space difference measurement and is used to guide the model to learn the mapping relationship from the medium-free state to the target enhanced state.

[0133] The image enhancement model described above is an initial enhancement model after parameter adjustment. Its internal parameters have been improved by minimizing the image loss function, enabling it to generate a simulated enhanced image that is highly consistent with the visual perception and structural distribution of the sample image when a new second sample image is input.

[0134] In one optional embodiment, a second sample image to be processed and its corresponding sample developed image can be obtained to construct a paired training sample set. Next, the second sample image to be processed is predicted using the initial enhancement model to generate a predicted enhanced image that simulates the enhancement effect of the target medium. Then, the differences between the predicted enhanced image and the sample developed image in terms of structure, contrast, and texture dimensions are calculated to construct an image loss function. Finally, based on the gradient information of the image loss function, the parameters of the initial enhancement model are iteratively adjusted so that the model output gradually approximates the distribution characteristics of the sample developed image, forming an image enhancement model with generalization ability.

[0135] The system can acquire a second sample image to be processed and a corresponding sample development image, where the sample development image is generated after injecting the target medium into at least one biological tissue structure, reflecting the contrast enhancement morphology of the target region. An initial enhancement model is used to predict the second sample image to be processed, simulating the visual changes after target medium injection through multi-layer feature mapping and spatial transformation to generate a predicted enhanced image. Based on the sample development image and the predicted enhanced image, the differences in overall structure and grayscale distribution are calculated, and an image loss function is constructed. This function guides the model to gradually approximate the contrast enhancement pattern presented by the sample development image. Based on the image loss function, the model parameters of the initial enhancement model are iteratively adjusted, enabling the model to learn the mapping rules from an unenhanced image to a developed image without relying on real medium injection, ultimately obtaining an image enhancement model. This model can generate a predicted enhanced image with similar contrast enhancement effects on an input image without target medium injection, improving the recognizability and consistency of image information.

[0136] This application's embodiments introduce a sample development image as a supervisory signal for the true enhancement state, and combine it with an image loss function to guide the model to learn a nonlinear mapping from a medium-free image to a target enhanced image. This enables the image enhancement model to automatically generate virtual enhancement results that conform to physiological and physical laws without the need for manual annotation of enhancement regions, thereby achieving effective simulation of contrast agent-dependent imaging.

[0137] This application can perform image segmentation on the input image to obtain a mask image of the detection region in the image. It reconstructs the high-contrast structural features of the detection region without external enhancement intervention, making it suitable for scenarios with limited or resource-constrained enhancement intervention, and reducing reliance on real enhancement data. The spatial positioning information represented by the mask image of the detection region can be used as an explicit conditional input, guiding the generation process to focus on the three-dimensional distribution state of the detection region, improving the topological consistency of the output structure and the reliability of the generation rules, supporting the discrimination of key region features, and reducing the false detection rate and improving the interpretation confidence in the screening of the image to be processed.

[0138] Figure 3 This is a schematic diagram of an image processing procedure according to an embodiment of this application, such as... Figure 3 As shown, a computed tomography (CT) image and Gaussian noise can be input into a diffusion model (Conditional diffusion model). A segmentation model can be used to segment the CT image to obtain a mask image of the detection region in the CT image. The mask image is then input into the diffusion model to determine the spatial location information of the detection region in the image to be processed. The spatial location information is then input into the diffusion model, and the diffusion model is guided to iteratively denoise the input CT image and Gaussian noise based on the spatial location information to obtain an initial enhanced CT image. The initial enhanced CT image is then magnified to a preset resolution to obtain a magnified image. Based on the preset resolution, detail restoration is performed on the magnified image to obtain the target enhanced CT image.

[0139] This application uses the tumor and peripancreatic vascular mask as explicit spatial conditions to drive a 3D conditional diffusion model to accurately generate a virtual contrast enhancement effect that conforms to physiological laws during the reverse denoising process. Furthermore, it enhances detail clarity through a cascaded super-resolution network, achieving stable and anatomically consistent synthesis from non-enhanced CT to high-fidelity virtual CE-CT, effectively overcoming the artifact and blurring problems of traditional methods. This method requires no manual intervention, can be fully automated, and is suitable for scenarios involving iodine allergies or contrast agent shortages. The generated images closely approximate real CE-CT in terms of vascular continuity, tumor enhancement dynamics, and noise texture, improving the assessment of pancreatic cancer resectability and controlling the rate of missed diagnoses.

[0140] This application discloses a method and system for 3D volume data augmentation based on generative artificial intelligence. It employs a three-stage cascaded automatic generation pipeline to automatically generate high-resolution target computed tomography-enhanced images from low-contrast computed tomography images. This pipeline decouples low-frequency semantic information that needs to follow structural distribution from high-frequency realistic details such as edges, textures, and noise in stages. Furthermore, it imposes spatial constraints on the generation process by detecting spatial localization information represented by the region mask image, thereby reducing the risk of artifacts and non-physiological enhancements.

[0141] In one optional embodiment, when the contrast of the image to be processed is low, key biological tissue structures requiring focused enhancement, i.e., the detection areas mentioned above, can be automatically identified and located to provide accurate spatial priors for subsequent generation. The image to be processed contains three-dimensional volume data, and preprocessing operations can be performed on the image, such as voxel spacing resampling, intensity normalization, and target region cropping. Subsequently, a three-dimensional segmentation network is invoked to automatically segment the target biological tissue structure and its associated spatial structures, outputting a voxel-level mask image (which can be binary, multi-class, or probabilistic). This voxel-level mask image can contain the main structure of at least one biological tissue structure and its adjacent associated structures.

[0142] When training the above-mentioned 3D segmentation network, the mask image can be derived from expert annotation or an iterative annotation process of "pseudo-label + manual correction". The mask image can be used as an explicit condition input into the subsequent processing model to guide the enhancement range, suppress abnormal disturbances in non-target regions, and improve the stability of structural continuity and spatial relationship expression.

[0143] Low-frequency structural remapping can be achieved using a diffusion model. While maintaining the topological consistency of biological tissue structures, the mapping pattern from "low-contrast input to high-contrast output" can be learned to generate virtual augmentation data that conforms to the target phase feature distribution—the initial augmented image mentioned above. To balance computational efficiency and memory usage, the input 3D volume data can be downsampled in the axial plane, adjusting from high resolution to medium resolution, and then performing 3D generation calculations at this resolution. During the generation process, initialization can begin with random noise, and denoising and content refinement operations can be progressively performed through multiple iterations, ultimately converging to the virtual augmentation data of the target phase to obtain the initial augmented image. The conditional input can consist of two parts: the image to be processed and a mask image of the detection region. The image to be processed provides the individual structural basis and density distribution prior, while the mask image provides display space constraints.

[0144] In the feature modulation stage, a spatial adaptive adjustment mechanism can be introduced so that the mask image can guide the generation process in depth rather than just be a simple stitch. The mask image can be encoded into spatial guidance parameters of the same scale as the intermediate feature map, and the feature response at different spatial locations can be differentially modulated. This allows the diffusion model to learn enhancement patterns that conform to the target distribution in the detection region boundary, internal structure and adjacent associated regions, while suppressing non-physiological enhancements in cavities, low-density regions and other regions that are susceptible to registration perturbations.

[0145] To improve the stability of training and inference with large-volume data, a dynamic dual-path pre-mechanism can be adopted. The network can output two types of information in parallel: the denoising residual term required by the traditional diffusion process, and the structural content term that is closer to the target enhanced image itself. Learnable fusion weights can be introduced to adaptively balance the contributions of the two, thereby simultaneously achieving structural stability and texture realism in different regions. Among them, structural stability is mainly used to avoid topological deformation and anatomical distortion, while texture realism is mainly used to avoid blurring, edge blunting, and loss of details. To reduce the risk of the model learning "pseudo-enhancement patterns" due to registration errors, self-regularization constraints can be introduced to apply local consistency penalties to regions prone to spatial alignment deviations, preventing the model from learning registration errors as valid structural changes. Based on the aforementioned model processing, an initial enhanced image at medium resolution can be output. By accurately restoring the contrast distribution pattern of the detection region, the structural topological continuity and spatial relationship rationality can be guaranteed, providing a high-fidelity foundation for high-frequency detail restoration.

[0146] Furthermore, high-frequency detail reconstruction can be performed on the initial enhanced image. By restoring the high-frequency details of the detection region in the initial enhanced image, the final generated target enhanced image can more closely resemble the real enhanced sample in terms of edge sharpness, structural contour continuity, texture representation, and noise distribution characteristics, thereby improving visual readability and human-computer interaction experience. Specifically, the output medium-resolution initial enhanced image can be sliced ​​axially layer by layer and input into a two-dimensional image enhancement model to perform super-resolution reconstruction (e.g., restoring 256×256 to 512×512, achieving a two-fold magnification). The image enhancement model used can be an adversarial generative structure, which is better at restoring local texture details and perceptual realism. At the same time, a random noise injection mechanism is introduced into the generator to simulate the inherent noise characteristics of the original acquisition process, avoiding excessive smoothing and texture loss in the output.

[0147] A hybrid training strategy can be used to train an image enhancement model. The low-resolution input samples contain two types of data: an initial enhanced image of medium resolution and a low-resolution sample obtained by downsampling a real high-resolution enhanced image. During downsampling, it is necessary to ensure that the image's pixel-level alignment with spatial coordinates is maintained to stably learn the detail recovery rules. This hybrid training strategy can simultaneously consider the diversity of real-world generated distributions and the accuracy of real-world alignment structures during training. This allows the model to maintain 3D structural stability during inference while accurately enhancing the boundary sharpness, contour continuity, internal texture, and contrast of neighboring regions of the detection area, outputting a final high-resolution target enhanced image. The target enhanced image can be saved in a standard data format, inheriting key metadata (such as window width, window level, voxel spacing, etc.) from the original image to be processed, ensuring seamless compatibility with downstream display systems and workstation platforms. It can be used directly for observation and interpretation without additional configuration.

[0148] This application employs a systematic architecture of mask prior constraints, 3D conditional diffusion generation, and cascaded detail enhancement. In terms of generation mechanism, it replaces the traditional model with a diffusion model, utilizing a multi-step iterative denoising generation process to improve the stability and texture fidelity of the output structure, effectively suppressing artifacts and non-physiological enhancements caused by adversarial training. Simultaneously, the diffusion model performs end-to-end modeling in the 3D volumetric data space, improving the structural continuity and cross-layer consistency of the target region in the depth direction. Furthermore, this application introduces a detection region mask image automatically generated in the preceding stage as an explicit spatial constraint, and deeply embeds it into the generation process through a spatial adaptive modulation mechanism. This allows the diffusion model to learn physiologically appropriate enhancement responses in the detection region and its associated structures, while suppressing abnormal perturbations in non-target regions.

[0149] To address potential errors in multi-phase registration in low-density regions, this application introduces self-regularization constraints during the generation stage. This imposes local consistency penalties on mismatch-prone regions, preventing the diffusion model from misjudging registration deviations as valid structural changes, thereby improving the overall structural topological reliability. Finally, to mitigate the inherent smoothing tendency of the diffusion model in high-frequency detail recovery, this application adds a super-resolution enhancement stage. Combining an adversarial enhancement model with a noise injection mechanism, it accurately restores the edge sharpness, contour continuity, and texture graininess of the target region, making the final output target-enhanced image visually and interpretably close to the real enhanced sample.

[0150] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0151] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0152] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, they can also be implemented by hardware. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0153] According to an embodiment of this application, an image processing method is provided. Figure 4 This is a flowchart of another image processing method according to an embodiment of this application, such as... Figure 4 As shown, the method includes:

[0154] Step S402: In response to the input command applied to the operation interface, display the image to be processed on the operation interface.

[0155] The image to be processed contains a detection area containing at least one biological tissue structure.

[0156] After receiving input instructions from the operation interface, the image to be processed can be loaded and displayed. The image to be processed is three-dimensional volume data of the uninjected target medium, which contains a detection area of ​​at least one biological tissue structure. The shape of the area is blurred due to the low contrast between biological tissues, but the structural position is completely preserved by the original data.

[0157] Step S404: In response to the processing command applied to the operation interface, display the target enhancement image on the operation interface.

[0158] Among them, the target enhancement image is obtained by restoring details of the initial enhancement image based on a preset resolution. The initial enhancement image is obtained by denoising the image to be processed based on spatial positioning information. The spatial positioning information is obtained based on the mask image of the detection area. The mask image is obtained by image segmentation of the image to be processed. The display clarity of the detection area in the target enhancement image is higher than that of the detection area in the image to be processed. The spatial positioning information is used to represent the distribution state of the detection area in the corresponding three-dimensional space of the image to be processed.

[0159] After receiving processing instructions from the operation interface, the image enhancement process is executed. First, a mask image is generated from the image to be processed through image segmentation to identify the spatial range of the detection area. Then, spatial positioning information is extracted based on the mask image as a structural constraint for denoising. Subsequently, the original image is denoised based on the spatial positioning information to output the initial enhanced image. Finally, the initial enhanced image is restored with a preset resolution to generate the target enhanced image, making the boundaries and structure of the detection area clearer in the display. The improvement in clarity comes from spatially guided denoising and resolution-controlled texture reconstruction, without introducing non-realistic enhancement information.

[0160] Through the above steps, in response to input commands applied to the operation interface, the image to be processed is displayed on the operation interface, wherein the image to be processed contains at least one detection region of a biological tissue structure; in response to processing commands applied to the operation interface, a target enhancement image is displayed on the operation interface, wherein the target enhancement image is obtained by detail restoration of the initial enhancement image based on a preset resolution, the initial enhancement image is obtained by denoising the image to be processed based on spatial positioning information, the spatial positioning information is obtained based on a mask image of the detection region, the mask image is obtained by image segmentation of the image to be processed, the display clarity of the detection region in the target enhancement image is higher than the display clarity of the detection region in the image to be processed, and the spatial positioning information is used to indicate the location of the detection region in the image to be processed. Corresponding to the distribution state in three-dimensional space, the goal of improving image processing effect is achieved. The mask image obtained by image segmentation of the image to be processed represents the spatial distribution state of the detection area in the image to be processed. Based on this spatial positioning information, selective denoising is performed on the image to be processed, so that the denoising process only acts on the spatial position corresponding to the detection area, thereby preserving the original characteristics of the non-target area. The initial enhanced image obtained after denoising has higher structural consistency in the detection area. Then, according to the preset resolution, the initial enhanced image is subjected to detail restoration, so that the display clarity of the detection area is directionally improved under the constraint of spatial positioning information. Finally, the goal of the detection area in the target enhanced image is higher than that in the image to be processed is achieved, thereby solving the technical problem of poor image processing effect in related technologies.

[0161] According to an embodiment of this application, an image processing method is provided. Figure 5 This is a flowchart of another image processing method according to an embodiment of this application, such as... Figure 5 As shown, the method includes:

[0162] Step S502: Obtain the image to be processed by calling the first interface.

[0163] The first interface includes a first parameter, the value of which includes the image to be processed, and the image to be processed contains a detection region of at least one biological tissue structure.

[0164] The aforementioned first interface can be an interface for data interaction between the server and the client. The client can pass the image to be processed into the interface function as the first parameter of the interface function to achieve the purpose of uploading the image to be processed to the cloud server.

[0165] Step S504: Perform image segmentation on the image to be processed to obtain a mask image of the detection area.

[0166] Step S506: Based on the mask image, determine the spatial location information of the detection area in the image to be processed.

[0167] Among them, spatial positioning information is used to represent the distribution of the detection area in the three-dimensional space corresponding to the image to be processed.

[0168] Step S508: Denoise the image to be processed based on the spatial positioning information to obtain the initial enhanced image.

[0169] Step S510: Perform detail restoration on the initial enhanced image based on the preset resolution to obtain the target enhanced image.

[0170] Among them, the display clarity of the detection area in the target enhancement image is higher than that of the detection area in the image to be processed.

[0171] Step S512: Output the target enhanced image by calling the second interface.

[0172] The second interface includes a second parameter, the value of which includes the target enhancement image.

[0173] The second interface in the above steps can be an interface for data exchange between the cloud server and the client. The cloud server can pass the target augmented image into the interface function as the second parameter of the interface function to achieve the purpose of sending the target augmented image to the client.

[0174] Through the above steps, the image to be processed is obtained by calling a first interface, wherein the first interface includes a first parameter, the parameter value of which includes the image to be processed, and the image to be processed contains a detection region of at least one biological tissue structure; image segmentation is performed on the image to be processed to obtain a mask image of the detection region; based on the mask image, the spatial positioning information of the detection region in the image to be processed is determined, wherein the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; based on the spatial positioning information, denoising processing is performed on the image to be processed to obtain an initial enhanced image; detail restoration is performed on the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than the display clarity of the detection region in the image to be processed; the target enhanced image is output by calling a second interface. The enhanced image, wherein the second interface includes a second parameter, the parameter value of which includes the target enhanced image, thereby achieving the purpose of improving the image processing effect. The mask image obtained by image segmentation of the image to be processed represents the spatial distribution state of the detection area in the image to be processed. Based on this spatial positioning information, the image to be processed is selectively denoised, so that the denoising process only acts on the spatial position corresponding to the detection area, thereby preserving the original characteristics of the non-target area. The initial enhanced image obtained after denoising has higher structural consistency in the detection area. Then, according to the preset resolution, the initial enhanced image is subjected to detail restoration, so that the display clarity of the detection area is directionally improved under the constraint of the spatial positioning information. Finally, the goal of the detection area in the target enhanced image is higher than that in the image to be processed is achieved, thereby solving the technical problem of poor image processing effect in related technologies.

[0175] According to embodiments of this application, an image processing apparatus for implementing the above-described image processing method is also provided. Figure 6 This is a schematic diagram of an image processing apparatus according to an embodiment of this application, such as... Figure 6 As shown, the device 600 includes: a segmentation module 602, a determination module 604, a noise reduction module 606, and a recovery module 608.

[0176] The image segmentation module is used to segment the image to be processed to obtain a mask image of the detection region, wherein the image to be processed contains at least one detection region of biological tissue structure; the determination module is used to determine the spatial positioning information of the detection region in the image to be processed based on the mask image, wherein the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; the denoising module is used to denoise the image to be processed based on the spatial positioning information to obtain an initial enhanced image; the restoration module is used to restore details of the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than the display clarity of the detection region in the image to be processed.

[0177] It should be noted that the segmentation module 602, determination module 604, noise reduction module 606, and recovery module 608 correspond to steps S202 to S208 in the above embodiments. The four modules and their corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules or units can be hardware or software components stored in memory and processed by one or more processors. The above modules can also run as part of the device in the server 10 provided in the above embodiments.

[0178] In the above embodiments of this application, the restoration module is used to enlarge the initial enhanced image to a preset resolution to obtain an enlarged image; and to restore the details of the enlarged image using an image enhancement model to obtain a target enhanced image.

[0179] In the above embodiments of this application, the denoising module is used to control the denoising model to denoise the image to be processed based on spatial positioning information, so as to obtain an initial enhanced image.

[0180] In the above embodiments of this application, the denoising module is used to extract features from the image to be processed to obtain image texture features; to perform feature processing on spatial positioning information based on the feature scale of the image texture features to obtain spatial guidance features; and to perform denoising processing on the image to be processed based on the image texture features and spatial guidance features to generate an initial enhanced image.

[0181] In the above embodiments of this application, the denoising module is used to iteratively denoise the image to be processed based on spatial guided features to obtain a denoised image; to perform texture enhancement on the denoised image based on image texture features to obtain a texture-enhanced image; and to generate an initial enhanced image based on the denoised image and the texture-enhanced image.

[0182] In the above embodiments of this application, the device is further configured to acquire a first sample image to be processed, sample spatial positioning information of the first sample image to be processed, and a sample enhancement image corresponding to the first sample image to be processed; control an initial denoising model to denoise the first sample image to be processed based on the sample spatial positioning information to obtain a predicted enhancement image; construct a diffusion loss function based on the sample enhancement image and the predicted enhancement image; and adjust the model parameters of the initial denoising model based on the diffusion loss function to obtain a denoising model.

[0183] In the above embodiments of this application, the denoising module is used to extract features from the first sample image to be processed to obtain predicted image texture features; to perform feature processing on the sample spatial positioning information based on the feature scale of the predicted image texture features to obtain predicted spatial guidance features; to perform iterative denoising on the first sample image to be processed based on the predicted spatial guidance features to obtain a predicted denoised image; to perform texture enhancement on the predicted denoised image based on the predicted image texture features to obtain a predicted texture enhanced image; and to generate a predicted enhanced image based on the predicted denoised image and the predicted texture enhanced image.

[0184] In the above embodiments of this application, the denoising module is used to obtain the sample denoised image and the sample texture enhanced image corresponding to the first sample image to be processed; construct a denoising loss function based on the sample denoised image and the predicted denoised image; construct a texture enhancement loss function based on the sample texture enhanced image and the predicted texture enhanced image; and adjust the model parameters of the initial denoising model based on the diffusion loss function, the denoising loss function and the texture enhancement loss function to obtain the denoising model.

[0185] In the above embodiments of this application, the device is further configured to acquire a second sample image to be processed and a sample development image corresponding to the second sample image to be processed, wherein the sample development image is an image generated after injecting a target medium into at least one biological tissue structure; predict the second sample image to be processed using an initial enhancement model to obtain a predicted enhancement image; construct an image loss function based on the sample development image and the predicted enhancement image; and adjust the model parameters of the initial enhancement model based on the image loss function to obtain an image enhancement model.

[0186] According to embodiments of this application, an image processing apparatus for implementing the above-described image processing method is also provided. Figure 7 This is a schematic diagram of another image processing apparatus according to an embodiment of this application, such as... Figure 7 As shown, the device 700 includes: a first display module 702 and a second display module 704.

[0187] The first display module is used to respond to input commands applied to the operation interface and display the image to be processed on the operation interface. The image to be processed contains at least one detection area of ​​biological tissue structure. The second display module is used to respond to processing commands applied to the operation interface and display the target enhancement image on the operation interface. The target enhancement image is obtained by restoring details of the initial enhancement image based on a preset resolution. The initial enhancement image is obtained by denoising the image to be processed based on spatial positioning information. The spatial positioning information is obtained based on a mask image of the detection area. The mask image is obtained by image segmentation of the image to be processed. The display clarity of the detection area in the target enhancement image is higher than that of the detection area in the image to be processed. The spatial positioning information is used to indicate the distribution state of the detection area in the corresponding three-dimensional space of the image to be processed.

[0188] It should be noted that the first display module 702 and the second display module 704 mentioned above correspond to steps S402 to S404 in the above embodiments. The two modules and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules or units can be hardware components or software components stored in memory and processed by one or more processors. The above modules can also be part of the device and run in the server 10 provided in the above embodiments.

[0189] According to embodiments of this application, an image processing apparatus for implementing the above-described image processing method is also provided. Figure 8 This is a schematic diagram of another image processing apparatus according to an embodiment of this application, such as... Figure 8 As shown, the device 800 includes: an acquisition module 802, a segmentation module 804, a determination module 806, a noise reduction module 808, a recovery module 810, and an output module 812.

[0190] The system comprises the following modules: an acquisition module for acquiring an image to be processed by calling a first interface, wherein the first interface includes a first parameter whose value includes the image to be processed, and the image to be processed contains a detection region of at least one biological tissue structure; a segmentation module for segmenting the image to be processed to obtain a mask image of the detection region; a determination module for determining the spatial location information of the detection region in the image to be processed based on the mask image, wherein the spatial location information represents the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; a denoising module for denoising the image to be processed based on the spatial location information to obtain an initial enhanced image; a restoration module for restoring details of the initial enhanced image based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection region in the target enhanced image is higher than that of the detection region in the image to be processed; and an output module for outputting the target enhanced image by calling a second interface, wherein the second interface includes a second parameter whose value includes the target enhanced image.

[0191] It should be noted that the acquisition module 802, segmentation module 804, determination module 806, noise reduction module 808, recovery module 810, and output module 812 mentioned above correspond to steps S502 to S512 in the above embodiments. The six modules and their corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules or units can be hardware or software components stored in memory and processed by one or more processors. The above modules can also run as part of the device in the server 10 provided in the above embodiments.

[0192] According to an embodiment of this application, a computer-aided diagnosis method for tumors is provided. Figure 9 This is a flowchart of a computer-aided diagnosis method for tumors according to an embodiment of this application, such as... Figure 9 As shown, the method includes:

[0193] Step S902: Perform image segmentation on the computed tomography image to obtain a mask image of the detection area.

[0194] Computed tomography (CT) images contain areas for detecting biological tissues and tumors.

[0195] The aforementioned biological tissues may be pancreatic tissue, liver tissue, kidney tissue, etc., and this application uses pancreatic tissue as the example for explanation. The aforementioned tumors may be pancreatic tumors, liver tumors, kidney tumors, etc., and this application uses pancreatic tumors as the example for explanation.

[0196] In one alternative embodiment, deep learning segmentation can be performed on non-enhanced CT images to automatically identify and accurately delineate target areas such as pancreatic tumors, pancreatic parenchyma, and surrounding key blood vessels, generating high-precision three-dimensional mask images as the structural basis for subsequent generation processes, ensuring that the model focuses on key clinical anatomical structures.

[0197] Step S904: Based on the mask image, determine the spatial location information of the detection area in the computed tomography image.

[0198] Among them, spatial positioning information is used to represent the distribution of the detection area in the three-dimensional space corresponding to the scanned image in the computer segment.

[0199] In one alternative embodiment, the location, shape, and topological relationship of tumors and blood vessels in three-dimensional space can be extracted based on the segmented mask to form structured spatial positioning information, providing clear anatomical guidance for the generated model, avoiding virtual enhancement from deviating from the real physiological distribution, and improving the anatomical consistency of the generated results.

[0200] Step S906: Denoise the computed tomography (CT) image based on spatial positioning information to obtain an initial enhanced CT image.

[0201] In one alternative embodiment, spatial positioning information can be used as a conditional input to drive a three-dimensional conditional diffusion model to perform a reverse denoising process. While preserving the original CT anatomical structure, contrast enhancement features that conform to physiological laws are gradually injected to simulate the dynamic uptake of contrast agents in tumors and blood vessels during the arterial and venous phases, thereby generating preliminary virtual CE-CT images.

[0202] Step S908: Perform detail restoration on the initial computed tomography enhanced scan image based on a preset resolution to obtain the target computed tomography enhanced scan image.

[0203] In one alternative embodiment, to address the potential detail blurring issue in images generated by the diffusion model, a cascaded super-resolution network can be used to perform layer-by-layer detail restoration on axial slices, enhancing the contours of blood vessel walls, the sharpness of tumor edges, and the inherent noise texture of CT, so that the images are close to the high-resolution CE-CT acquired in clinical settings in terms of resolution and realism.

[0204] Step S910: Perform auxiliary diagnosis of the tumor based on the target computed tomography enhanced scan image to obtain auxiliary diagnostic results.

[0205] In one alternative embodiment, the resulting high-fidelity virtual CE-CT images can be integrated into the clinical diagnostic process to assist radiologists in identifying tumor boundaries, assessing the extent of vascular invasion, determining resectability, and reducing misdiagnosis and missed diagnosis caused by low contrast, thereby achieving accurate auxiliary diagnosis and decision support under contrast agent-free conditions.

[0206] This application can automatically generate high-fidelity, anatomically consistent virtual contrast-enhanced CT (virtual CE-CT) images from non-contrast computed tomography (NC-CT) images, achieving accurate tumor assessment without contrast agents. This application includes several computer-executed steps. First, a deep learning model performs three-dimensional voxel segmentation on the NC-CT image to accurately delineate the tumor and key tissues (such as the pancreas and peripancreatic vessels), generating a structured mask, i.e., the aforementioned mask image, to provide clear anatomical guidance for subsequent processing. Second, based on this mask image, the geometric morphology, spatial coordinates, and topological relationships of the tumor and blood vessels in three-dimensional space are extracted to form precise spatial positioning information, ensuring that the enhancement process strictly follows the logic of real anatomy, eliminating illusions and misalignments. This positioning information is used as a conditional input to drive a three-dimensional diffusion model to perform reverse denoising on the computed tomography image, simulating the physiological enhancement dynamics of contrast agents in the tumor and vascular system. Without pre-setting a time curve, an initial computed tomography enhanced image with arterial and venous phase characteristics can be generated solely from learned real CE-CT data.

[0207] To improve the clinical usability of the images, a cascaded super-resolution network is used to restore axial slice-level details of the preliminary results, improving the resolution to the clinical standard of 0.5–1.0 mm. It accurately restores inherent CT noise, vascular wall texture, and tumor edge sharpness, making the generated images highly approximate real CE-CT in terms of visual realism, structural detail, and noise distribution—that is, the target computed tomography-enhanced scan image. The final target computed tomography-enhanced scan image is directly embedded into the radiology diagnostic workflow to assist physicians in determining tumor resectability, the degree of vascular invasion, and benign / malignant characteristics, improving the detection rate of low-contrast lesions and reducing false positive and false negative rates. This application eliminates the need for contrast agent injection, making it suitable for cancer patients with iodine allergy, renal insufficiency, and those requiring long-term follow-up. It not only reduces clinical risks and contrast agent dependence but also promotes the evolution of radiology towards safety, intelligence, and sustainability, possessing clear clinical translational value and broad applicability.

[0208] Through the above steps, image segmentation is performed on the computed tomography (CT) image to obtain a mask image of the detection region. The CT image contains detection regions for biological tissue and tumors. Based on the mask image, the spatial location information of the detection region in the CT image is determined. This spatial location information represents the distribution of the detection region in the corresponding three-dimensional space of the CT image. Based on the spatial location information, the CT image is denoised to obtain an initial enhanced CT image. Details are restored from the initial enhanced CT image at a preset resolution to obtain a target enhanced CT image. Based on the target enhanced CT image, auxiliary diagnosis of the tumor is performed to obtain... The auxiliary diagnostic results improve the image processing effect. The mask image obtained by image segmentation represents the spatial distribution of the detection area in the image to be processed. Based on this spatial positioning information, selective denoising is performed on the image to be processed, so that the denoising process only acts on the spatial position corresponding to the detection area, thereby preserving the original characteristics of the non-target area. The initial enhanced image obtained after denoising has higher structural consistency in the detection area. Then, according to the preset resolution, the initial enhanced image is subjected to detail restoration, so that the display clarity of the detection area is directionally improved under the constraint of spatial positioning information. Finally, the goal of the detection area in the target enhanced image is higher than that in the image to be processed is achieved, thereby solving the technical problem of poor image processing effect in related technologies.

[0209] It should be noted that the preferred embodiments involved in the above embodiments of this application are the same as the solutions, application scenarios and implementation processes provided in the above embodiments, but are not limited to the solutions provided in the above embodiments.

[0210] Embodiments of this application may provide a computing device. Figure 10 This is a schematic diagram of another image processing apparatus according to an embodiment of this application. Figure 10 As shown, the computing device 100 may include one or more (only one is shown in the figure) processors 102, memory 104, memory controller, and peripheral interfaces.

[0211] The aforementioned computing device can be understood as an integrated smart terminal, including but not limited to servers, desktop computers, PCs (Personal Computers), all-in-one model machines, etc., and the computing device may have the model in the above embodiments of this application pre-installed.

[0212] Specifically, this computing device can pre-install various types of models, including but not limited to models in natural language processing, visual processing, speech processing, code processing, and multimodal task processing, thus providing diverse model selection. In different product forms, this computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine-tuning, model deployment, model inference, and application. In some product forms, this computing device also supports model management, including but not limited to multi-type model management (supporting the management of discriminative, generative, and other types of models), model version control (supporting the control of different model versions), and model evaluation (evaluating model performance and effectiveness based on model evaluation tools). In other product forms, this computing device can also create applications based on models, providing API calling capabilities, allowing models to be called into created applications through API interfaces, and providing application management tools for application management and monitoring.

[0213] Furthermore, the computing device can also include data management (supporting the creation and management of model tuning datasets), a training center (providing abundant training resources to help users learn and master AI technology), and basic control capabilities (providing enterprise-level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, it provides a comprehensive and integrated device for AI development, training, deployment, and application.

[0214] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the methods in the above embodiments. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to terminal A via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0215] The processor can invoke an executable program stored in memory via a transmission device to execute any of the methods described in the above embodiments.

[0216] Embodiments of this application may provide an electronic device. Figure 11 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 11 As shown, the electronic device may include: an input / output device 112; a memory 114; and a processor 116, wherein the processor 116 is connected to the input / output device 112 and the memory 114 via a bus 118.

[0217] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the methods in the above embodiments. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to terminal A via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0218] The processor can invoke an executable program stored in memory via a transmission device to execute any of the methods described in the above embodiments.

[0219] Those skilled in the art will understand that, Figure 11 The structure shown is for illustrative purposes only. The computing device may also be a smartphone, tablet, PDA, mobile internet device (MID), PAD, or other terminal device. This figure does not limit the structure of the computing device described above. For example, computing device 100 may include more or fewer components (such as network interfaces, display devices, etc.) than shown in this figure, or may have a different configuration than shown in this figure.

[0220] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0221] Embodiments of this application also provide a computer-readable storage medium. Optionally, in this embodiment, the computer-readable storage medium can be used to store program code executed by the method provided in the above embodiments.

[0222] Optionally, in this embodiment, the storage medium may be located in a computing device.

[0223] Optionally, in this embodiment, the computer-readable storage medium is configured to store an executable program, which, when the executable program is running, controls the device where the computer-readable storage medium is located to execute the method described in any of the above embodiments.

[0224] Embodiments of this application also provide a computer program product. Optionally, in this embodiment, the computer program product may include a computer program that, when executed by a processor, implements the methods provided in the embodiments described above.

[0225] Embodiments of this application also provide a computer program product. Optionally, the computer program product may include a non-volatile computer-readable storage medium, which can be used to store a computer program that, when executed by a processor, implements the method provided in the above embodiments.

[0226] Embodiments of this application also provide a computer program. Optionally, in this embodiment, when the computer program is executed by a processor, it implements the method provided in the above embodiments.

[0227] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0228] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0229] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0230] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0231] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0232] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An image processing method, characterized in that, include: Image segmentation is performed on the image to be processed to obtain a mask image of the detection region, wherein the image to be processed contains the detection region containing at least one biological tissue structure; Based on the mask image, the spatial positioning information of the detection region in the image to be processed is determined, wherein the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; Based on the spatial positioning information, the image to be processed is denoised to obtain an initial enhanced image; The initial enhanced image is subjected to detail restoration based on a preset resolution to obtain a target enhanced image, wherein the display clarity of the detection area in the target enhanced image is higher than that of the detection area in the image to be processed.

2. The method according to claim 1, characterized in that, Based on a preset resolution, detail restoration is performed on the initial enhanced image to obtain the target enhanced image, including: The initial enhanced image is magnified to the preset resolution to obtain the magnified image; The magnified image is then subjected to detail restoration using an image enhancement model to obtain the target enhanced image.

3. The method according to claim 2, characterized in that, Based on the spatial positioning information, the image to be processed is denoised to obtain an initial enhanced image, including: Based on the spatial positioning information, the denoising model is controlled to perform denoising processing on the image to be processed, and an initial enhanced image is obtained; The process involves using a denoising model controlled by the spatial positioning information to denoise the image to be processed, resulting in an initial enhanced image, including: Feature extraction is performed on the image to be processed to obtain image texture features; Based on the feature scale of the image texture features, feature processing is performed on the spatial positioning information to obtain spatial guidance features; The image to be processed is denoised based on the image texture features and the spatial guidance features to generate the initial enhanced image.

4. The method according to claim 3, characterized in that, Based on the image texture features and the spatial guidance features, the image to be processed is denoised to generate the initial enhanced image, including: Based on the spatial guidance features, the image to be processed is iteratively denoised to obtain a denoised image; Based on the image texture features, the denoised image is enhanced to obtain a texture-enhanced image; The initial enhanced image is generated based on the denoised image and the texture enhanced image.

5. The method according to claim 3, characterized in that, The method further includes: Obtain the first sample image to be processed, the sample spatial positioning information of the first sample image to be processed, and the sample enhancement image corresponding to the first sample image to be processed; Based on the sample spatial positioning information, the initial denoising model is controlled to denoise the first sample image to be processed, thereby obtaining a predicted enhanced image. Based on the sample enhanced image and the predicted enhanced image, a diffusion loss function is constructed; The model parameters of the initial denoising model are adjusted based on the diffusion loss function to obtain the denoising model.

6. The method according to claim 5, characterized in that, Based on the sample spatial positioning information, the denoising model is controlled to denoise the first sample image to be processed, resulting in a predicted enhanced image, including: Feature extraction is performed on the first sample image to be processed to obtain the predicted image texture features; Based on the feature scale of the predicted image texture features, feature processing is performed on the sample spatial positioning information to obtain the predicted spatial guidance features; Based on the predicted spatial guided features, the first sample image to be processed is iteratively denoised to obtain a predicted denoised image; Based on the texture features of the predicted image, the predicted denoised image is texture-enhanced to obtain a predicted texture-enhanced image; The predicted enhanced image is generated based on the predicted denoised image and the predicted texture enhanced image.

7. The method according to claim 5, characterized in that, The model parameters of the initial denoising model are adjusted based on the diffusion loss function to obtain the denoising model, which includes: Obtain the denoised image and texture-enhanced image of the sample corresponding to the first sample image to be processed; Based on the sample denoised image and the predicted denoised image, a denoising loss function is constructed; Based on the sample texture enhancement image and the predicted texture enhancement image, a texture enhancement loss function is constructed; Based on the diffusion loss function, the denoising loss function, and the texture enhancement loss function, the model parameters of the initial denoising model are adjusted to obtain the denoising model.

8. The method according to claim 2, characterized in that, The method further includes: Acquire a second sample image to be processed and a corresponding sample development image, wherein the sample development image is an image generated after injecting the target medium into the at least one biological tissue structure; The initial enhancement model is used to predict the second sample image to be processed, resulting in a predicted enhanced image; Based on the developed sample image and the predicted enhanced image, an image loss function is constructed; The model parameters of the initial enhancement model are adjusted based on the image loss function to obtain the image enhancement model.

9. An image processing method, characterized in that, include: In response to an input command applied to the operation interface, an image to be processed is displayed on the operation interface, wherein the image to be processed contains a detection area of ​​at least one biological tissue structure. In response to a processing command applied to the operation interface, a target enhanced image is displayed on the operation interface. The target enhanced image is obtained by restoring details of an initial enhanced image based on a preset resolution. The initial enhanced image is obtained by denoising the image to be processed based on spatial positioning information. The spatial positioning information is obtained based on a mask image of the detection region. The mask image is obtained by image segmentation of the image to be processed. The display clarity of the detection region in the target enhanced image is higher than that of the detection region in the image to be processed. The spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed.

10. An image processing method, characterized in that, include: The image to be processed is obtained by calling a first interface, wherein the first interface includes a first parameter, the parameter value of the first parameter includes the image to be processed, and the image to be processed contains a detection area of ​​at least one biological tissue structure. The image to be processed is segmented to obtain a mask image of the detection region; Based on the mask image, the spatial positioning information of the detection region in the image to be processed is determined, wherein the spatial positioning information is used to represent the distribution state of the detection region in the corresponding three-dimensional space of the image to be processed; Based on the spatial positioning information, the image to be processed is denoised to obtain an initial enhanced image; Based on a preset resolution, detail restoration is performed on the initial enhanced image to obtain a target enhanced image, wherein the display clarity of the detection area in the target enhanced image is higher than the display clarity of the detection area in the image to be processed; The target enhanced image is output by calling a second interface, wherein the second interface includes a second parameter, and the parameter value of the second parameter includes the target enhanced image.

11. A computer-aided diagnostic method for tumors, characterized in that, include: Image segmentation is performed on a computed tomography (CT) image to obtain a mask image of the detection region, wherein the CT image contains the detection region containing biological tissue and tumor. Based on the mask image, the spatial positioning information of the detection area in the computed tomography image is determined, wherein the spatial positioning information is used to represent the distribution state of the detection area in the corresponding three-dimensional space of the computed tomography image; Based on the spatial positioning information, the computed tomography (CT) image is denoised to obtain an initial enhanced CT image. Based on a preset resolution, detail restoration is performed on the initial computed tomography enhanced scan image to obtain the target computed tomography enhanced scan image; The tumor is diagnosed using the enhanced computed tomography (CT) image of the target tumor, and the diagnostic results are obtained.

12. A computing device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 11.

13. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor, connected to the memory via a bus, is used to run the program, wherein the program, when running, executes the method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the computer-readable storage medium is located to perform the method according to any one of claims 1 to 11.

15. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 11.

Images