A rapid medical image data simulation method based on physical degradation modeling

Abstract

A rapid medical image data simulation method based on physical degradation modeling is proposed. First, 3D medical images are preprocessed to obtain basic medical image volume data suitable for projection simulation. Then, based on the scanning geometry parameters of the intraoperative cone-beam CT imaging system, the basic medical image volume data is forward-projected to obtain ideal projection data. Next, physical degradation factors are sequentially or selectively introduced into the projection domain to obtain simulated projection data with intraoperative cone-beam CT imaging degradation characteristics. Then, the simulated projection data is reconstructed into initial simulated medical image data using a 3D reconstruction method. Finally, combining structural prior information and a depth generation network, intraoperative imaging domain feature enhancement is performed on the initial simulated medical image data to output target simulated medical image data that possesses intraoperative cone-beam CT imaging characteristics and maintains anatomical structural consistency. This invention combines physical rationality, computational efficiency, parameter controllability, and data reproducibility.

Description

Technical Field

[0001] This invention belongs to the technical fields of medical image processing, medical image simulation, cone-beam CT imaging modeling and medical artificial intelligence training data construction, and particularly relates to a fast medical image data simulation method based on projection domain physical modeling and parametric artifact injection.

[0002] This invention can be used to rapidly generate simulated medical image data with intraoperative CBCT imaging features from preoperative CT, 4DCT or other three-dimensional medical images. It is applicable to scenarios such as intraoperative navigation, image-guided treatment, medical image registration, medical image reconstruction, cross-modal generative model training and medical artificial intelligence algorithm verification. Background Technology

[0003] With the development of intraoperative navigation, image-guided radiotherapy, interventional therapy, and orthopedic and thoracic surgeries, intraoperative CBCT imaging is increasingly used in clinical practice. Compared with preoperative diagnostic CT, intraoperative CBCT has advantages such as convenient acquisition and the ability to obtain real-time information about the patient's anatomical status during treatment or surgery. However, due to limitations in intraoperative acquisition conditions, scan dose, equipment structure, patient movement, scattering effects, and energy spectrum nonlinearity, intraoperative CBCT often suffers from problems such as high noise, insufficient soft tissue contrast, obvious streak artifacts, blurred edges, and incomplete structural representation. These problems can affect subsequent image registration, dose calculation, 3D reconstruction, navigation and positioning, and deep learning model training.

[0004] Current methods for acquiring medical imaging data primarily rely on real-world clinical data collection. However, under real-world clinical conditions, preoperative CT scans and intraoperative CBCT scans of the same patient often differ in acquisition time, patient position, respiratory status, and anatomical changes caused by intraoperative procedures, making it difficult to obtain strictly voxel-matched data. Furthermore, repeated acquisitions increase patient radiation dose and are also subject to clinical procedures and ethical constraints. Therefore, large-scale, high-quality, controllable, and structurally referenced medical imaging training data is difficult to obtain.

[0005] Existing medical image simulation methods can be broadly categorized as follows: The first category involves directly adding Gaussian noise, blurring, stripes, or grayscale transformations to the image domain. While simple to implement, these methods struggle to accurately reflect the complex artifacts caused by projection inconsistencies, beam hardening, scattering, and photon statistical noise during CBCT imaging. The second category comprises methods based on Monte Carlo or complete physical system modeling. These methods offer high simulation accuracy but suffer from high computational complexity and time consumption, making them unsuitable for rapid generation of large-scale training data. The third category consists of cross-modal generation methods based on deep learning. These methods rely on a large amount of real-world paired or approximately paired data, and their training stability and generalization ability remain limited when real medical data is insufficient.

[0006] Therefore, a fast, controllable, and reproducible method for simulating medical image data is needed. This method should be able to simulate common motion artifacts, beam hardening artifacts, scattering background, and photon statistical noise in intraoperative CBCT using existing CT or 4DCT data at a relatively low computational cost, and then reconstruct three-dimensional medical image data with realistic intraoperative features. This method can not only provide training samples for medical image registration, reconstruction, and cross-modal generation models, but also provide a simulation data foundation with a well-defined structural reference for algorithm performance evaluation. Summary of the Invention

[0007] To overcome the shortcomings of existing medical image data simulation methods, this invention provides a rapid medical image data simulation method, aiming to solve the following technical problems: First, there is the difficulty in obtaining strictly paired preoperative CT and intraoperative CBCT data in real clinical scenarios. Due to differences in acquisition time, patient position, respiratory status, scanning equipment, and intraoperative operating conditions between preoperative CT and intraoperative CBCT, it is difficult to obtain large-scale, strictly corresponding, and reusable training and validation data.

[0008] Second, existing empirical simulation methods in the image domain lack realism. Traditional methods typically add noise, blur, or grayscale disturbances directly to the image domain, making it difficult to accurately reflect the complex artifacts caused by motion, beam hardening, scattering background, and photon statistical noise during CBCT imaging.

[0009] Third, existing high-precision physical simulation methods suffer from high computational costs. While methods based on complete energy spectrum integration, Monte Carlo scattering simulation, or complex biomechanical deformation modeling offer high physical realism, their high computational complexity and slow generation speed make them unsuitable for constructing large-scale medical image training data.

[0010] Fourth, there is a lack of controllable simulation data for medical image registration, reconstruction, and cross-modal generation models. Existing data often lack clear structural references, making it difficult to quantitatively compare and interpret different algorithms under the same imaging conditions.

[0011] Therefore, the purpose of this invention is to provide a rapid medical image data simulation method that combines physical rationality, computational efficiency, parameter controllability, and data reproducibility, for rapidly generating simulated medical image data with intraoperative CBCT imaging features from preoperative CT, 4DCT, or other three-dimensional medical images.

[0012] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A rapid medical image data simulation method based on physical degradation modeling is proposed. First, three-dimensional medical image data is acquired and standardized preprocessed to obtain basic medical image volume data. Second, based on the scanning geometry parameters of the intraoperative cone-beam CT imaging system, the basic medical image volume data is forward-projected to obtain ideal projection data. Then, based on the ideal projection data, intraoperative imaging-related physical degradation factors are introduced, including at least one of motion artifacts, beam hardening artifacts, photon statistical noise, and scattering effects, to generate simulated projection data with intraoperative imaging degradation characteristics. Next, the simulated projection data is reconstructed in three dimensions to obtain initial simulated medical image data. Subsequently, structural prior information is constructed based on the basic medical image data and the initial simulated medical image data, and the basic medical image data, the initial simulated medical image data, and the structural prior information are input into a depth generation network for intraoperative imaging domain feature enhancement. Finally, target simulated medical image data with intraoperative cone-beam CT imaging characteristics and maintaining anatomical structural consistency is output.

[0013] Furthermore, the method includes the following steps: 1) Image acquisition and preprocessing: Acquire preoperative CT or other 3D medical image data; perform spatial resampling, grayscale normalization, cropping, registration or region of interest extraction on the acquired data to obtain basic volume data in a unified coordinate system; 2) Based on the geometric parameters of the target intraoperative cone-beam CT imaging system, construct a cone-beam projection geometric model, and perform forward projection on the basic body data to obtain ideal projection data; 3) Introduce motion perturbation during or after the generation of the ideal projection data to obtain motion projection data containing motion artifact features; 4) Apply a material-dependent nonlinear attenuation mapping to the motion projection data to obtain beam-hardened projection data containing beam-hardening artifact features; 5) The beam-hardening projection data is converted to the intensity domain, and photon statistical noise and scattering components are introduced into the intensity domain. The simulated projection data is then obtained by normalization and logarithmic transformation. 6) Perform three-dimensional reconstruction on the simulated projection data to obtain simulated medical image data with intraoperative cone-beam CT imaging characteristics.

[0014] 7) Construct structural prior information based on the basic medical imaging data and / or the initial simulated medical imaging data; 8) The basic medical image data, initial simulated medical image data and structural prior information are used as input to the deep generation network; the deep generation network is used to enhance the features of the input image in the cone-beam CT imaging domain during surgery, so that the generated result has at least one feature among noise texture, scattering bias, beam hardening artifact, motion artifact, local contrast change or cone-beam CT image style. 9) Generate target simulated medical image data with intraoperative cone-beam CT imaging features.

[0015] In this invention, the input 3D medical image data is first standardized and preprocessed, including resampling, cropping, normalization, and attenuation coefficient conversion, to obtain basic medical image volume data suitable for forward projection. Then, a projection geometry model is constructed based on the intraoperative cone-beam CT system parameters, and projection simulation is performed on the basic medical image volume data. In the projection domain, the method sequentially introduces physical degradation factors such as motion perturbation, beam hardening, photon statistical noise, and scattering effects. After completing the degradation modeling, the simulated projection data is input into a 3D reconstruction algorithm to generate an initial simulated medical image. Furthermore, the method can construct structural prior information and input the basic image, the initial simulated image, and the structural prior into a depth generation network to enhance the domain features of the simulated image, making it closer to the noise texture, scattering bias, motion artifacts, beam hardening artifacts, and local contrast distribution of real intraoperative cone-beam CT while maintaining anatomical structural consistency.

[0016] The beneficial effects of this invention are as follows: This method can quickly generate simulated medical image data with intraoperative CBCT imaging characteristics and structural authenticity under the condition of lacking real and strictly paired preoperative CT and intraoperative CBCT data, which can be used for medical image registration, reconstruction, navigation and algorithm training and evaluation; it has the advantages of physical rationality, computational efficiency, parameter controllability and data reproducibility. Attached Figure Description

[0017] Figure 1 This is a flowchart of the overall process for rapid medical image data simulation based on physical degradation modeling.

[0018] Figure 2 This is a schematic diagram of the motion artifact simulation process in this invention.

[0019] Figure 3 This is a schematic diagram of the beam hardening artifact simulation process in this invention.

[0020] Figure 4 This is a schematic diagram of the simulation process for photon noise and scattering effects in this invention.

[0021] Figure 5 This is a flowchart illustrating the deep generation enhancement process of structural prior constraints in this invention. Detailed Implementation

[0022] The invention will now be further described with reference to the accompanying drawings.

[0023] Reference Figures 1-5 A rapid medical image data simulation method based on physical degradation modeling is proposed. First, pre-operative CT, 4DCT, or other 3D medical images are standardized and pre-processed, including resampling, intensity normalization, spatial alignment, and attenuation coefficient conversion, to obtain basic medical image volume data suitable for projection simulation. Then, based on the scanning geometry parameters of the intraoperative cone-beam CT imaging system, the basic medical image volume data is forward-projected to obtain ideal projection data. Next, physical degradation factors such as motion artifacts, beam hardening artifacts, photon statistical noise, and scattering effects are sequentially or selectively introduced into the projection domain to obtain simulated projection data with intraoperative cone-beam CT imaging degradation characteristics. Then, the simulated projection data is reconstructed into initial simulated medical image data using a 3D reconstruction method. Finally, combining structural prior information and a depth generation network, intraoperative imaging domain feature enhancement is performed on the initial simulated medical image data to output target simulated medical image data with intraoperative cone-beam CT imaging characteristics and maintaining anatomical structural consistency.

[0024] The method includes the following steps: 1) Image Acquisition and Preprocessing: Acquire preoperative CT, 4DCT, enhanced CT, or other three-dimensional medical image data of the object to be simulated, and preprocess the images. The preprocessing includes: voxel resampling of the original three-dimensional images to ensure that different samples meet a uniform spatial resolution; truncation and normalization of image grayscale to ensure that the tissue intensity distribution falls within a preset range; spatial alignment, region of interest cropping, and noise suppression of the images to reduce the impact of differences in the original acquisition on subsequent simulations; when the input image is CT or 4DCT, convert its CT values ​​into attenuation coefficient volume data suitable for line integral projection to ensure that the subsequent forward projection process matches the X-ray attenuation process.

[0025] 2) Projection geometry construction: Based on the scanning parameters of the target intraoperative cone-beam CT imaging system, a cone-beam projection geometry model is constructed; the scanning parameters include the position of the X-ray source, detector size, detector pixel spacing, source-to-detector distance, source-to-rotation center distance, scanning angle range, and number of projections; based on the projection geometry model, the basic medical image volume data is forward-projected to obtain ideal projection data, which is used as the basic input for subsequent physical degradation modeling.

[0026] 3) Motion artifact simulation: Refer to Figure 2Motion artifact simulation involves introducing motion perturbations during or after the generation of the ideal projection to simulate projection inconsistencies caused by patient movement, respiratory movement, organ movement, bed movement, equipment vibration, or sampling angle errors during intraoperative scanning. The motion artifact simulation process includes: setting the type, intensity, period, and proportion of motion perturbations based on the intraoperative scanning process; selecting a subset of angles from the projection angle sequence as motion perturbation angles and controlling their distribution within the scanning range; introducing sampling angle offsets to the motion perturbation angles and introducing spatial position offsets, rotational perturbations, or non-rigid deformations to the corresponding basic medical image volume data; subsequently, performing forward projection based on the perturbed sampling angles and the perturbed basic medical image volume data to obtain motion projection data containing motion inconsistencies. This step prevents the formation of motion artifact features such as stripes, ghosting, blurred edges, or local structural distortions in subsequent reconstructed images.

[0027] 4) Beam hardening artifact simulation: Refer to Figure 3 Based on motion projection data, beam hardening effects are further simulated. The beam hardening artifact simulation process includes: dividing the basic medical image volume data into bone tissue regions, soft tissue regions, air regions, or other material regions based on the grayscale values, CT values, tissue density, or tissue labels; performing projection calculations on different material regions to obtain the projection contributions corresponding to different materials; applying material-related nonlinear attenuation mappings to the projection contributions of each material based on the attenuation characteristics of different materials; and finally, synthesizing the nonlinear mapping results of different materials to obtain projection data with beam hardening artifact characteristics. Through this step, intensity bias, cupping artifacts, and high-density tissue periphery stripe artifacts caused by multi-energy spectral X-ray attenuation can be introduced into the simulation data.

[0028] 5) Simulation of photon noise and scattering effects: Refer to Figure 4 The projection data, after beam hardening modeling, is converted to the intensity domain, and photon statistical noise under low-dose acquisition conditions and common scattering effects in intraoperative cone-beam CT are simulated in the intensity domain. The photon noise and scattering effect simulation process includes: converting the projection data into corresponding X-ray intensity data; introducing photon statistical noise into the X-ray intensity data according to a preset dose level; constructing a low-frequency scattering background, spatially correlated scattering mask, or scattering kernel response, and adding them as scattering components to the intensity data; subsequently, normalizing and logarithmically transforming the intensity data after adding photon statistical noise and scattering components to obtain the final simulated projection data. This step can simulate common degradation phenomena in intraoperative cone-beam CT, such as non-stationary noise, low-frequency scattering bias, contrast degradation, and CT value drift.

[0029] 6) Generating Initial Simulated Images from 3D Reconstruction: The simulated projection data described above is input into a 3D reconstruction algorithm to obtain initial simulated medical image data. The 3D reconstruction method includes at least one of FDK analytical reconstruction, algebraic iterative reconstruction, statistical iterative reconstruction, prior constraint reconstruction, or deep learning reconstruction. In this embodiment, the FDK reconstruction method can be used to quickly obtain 3D simulated images with intraoperative cone-beam CT imaging characteristics. The initial simulated medical images retain the main anatomical structures of the basic medical images, while also including intraoperative imaging degradation features introduced by factors such as motion, beam hardening, noise, and scattering.

[0030] 7) Structural prior construction: Refer to Figure 5 Structural prior information is generated based on basic medical imaging data and / or initial simulated medical imaging data. This structural prior information includes at least one of the following: tissue partition masks, organ masks, edge maps, anatomical labels, deformation fields, multi-scale structural features, or structural label maps obtained by clustering methods. In one embodiment, unsupervised clustering of basic medical images can be performed using image grayscale or tissue density distribution to obtain major structural regions such as air, soft tissue, and bone tissue, which are then used as structural prior inputs to the depth generation network. By introducing structural priors, the consistency of anatomical structures can be constrained during subsequent enhancement processes, reducing structural shifts that occur during cross-modal or domain enhancement.

[0031] 8) Deep Generative Network Enhancement: Basic medical image data, initial simulated medical image data, and / or structural prior information are input into a deep generative network to enhance the intraoperative imaging domain features of the initial simulated medical images. The deep generative network can employ generative adversarial networks, recurrent consistent generative networks, diffusion models, Transformer networks, Mamba networks, selective state-space model networks, convolutional neural networks, or combinations thereof. The deep generative network may include an image input branch and a structural prior input branch. The image input branch is used to extract grayscale, texture, and artifact features, while the structural prior input branch provides tissue partitioning, boundary contours, and anatomical structure constraints. Furthermore, a long-range dependency modeling module can be introduced into the network to enhance the expressive power of global structure, edge details, texture distribution, and artifact features. Through this step, the generated results can more closely approximate the noise texture, scattering bias, beam hardening artifacts, motion artifacts, and local contrast distribution of real intraoperative cone-beam CT while maintaining anatomical structural consistency.

[0032] 9) Output Results: Output the target simulated medical image data and save the corresponding basic medical image, simulated projection data, projection geometry parameters, motion parameters, beam hardening parameters, noise parameters, scattering parameters, reconstruction parameters, and structural prior information. The target simulated medical image data can be used for medical image registration, image reconstruction, cross-modal generation, medical image segmentation, intraoperative navigation algorithm verification, data augmentation, or algorithm ablation experiments.

[0033] In this embodiment, by explicitly introducing multiple intraoperative imaging degradation factors into the projection domain, the formation process of the simulated data more closely approximates the actual cone-beam CT imaging mechanism. Compared with empirical simulation methods that directly superimpose noise or blurring into the image domain, this invention can more realistically express the structural degradation features caused by projection inconsistency, multi-spectral attenuation, low-dose sampling, and scattering background during scanning. Furthermore, since the simulated data is generated from basic medical images, these images can serve as a geometric reference standard, which is beneficial for subsequent registration, reconstruction, and training and validation of the generated model.

[0034] Reference Figure 1 The detailed mechanisms of each module are as follows: Data preprocessing module: This module is used to convert preoperative CT, 4DCT, enhanced CT or other three-dimensional medical images into standardized basic medical image volume data suitable for subsequent projection simulation and depth generation enhancement. It performs at least one of the following processing on the above medical image data: resampling, cropping, intensity normalization, spatial alignment, region of interest extraction or attenuation coefficient conversion, to obtain basic medical image volume data suitable for projection simulation. Physical Degradation Modeling Module: This module is used to construct a degradation model in the projection domain that approximates the actual intraoperative cone-beam CT imaging process. The physical degradation modeling module includes a motion artifact simulation unit, a beam hardening artifact simulation unit, a photon noise simulation unit, and a scattering effect simulation unit. Specifically, the motion artifact simulation unit perturbs the projection angle and anatomical structure position, causing different projection angles to correspond to different sampling geometries or anatomical states, thus creating projection inconsistencies; the beam hardening artifact simulation unit simulates intensity bias and fringe artifacts caused by multi-energy X-ray attenuation through material region partitioning and nonlinear attenuation mapping; the photon noise simulation unit simulates noise enhancement under low-dose acquisition through dose level control and stochastic statistical modeling; and the scattering effect simulation unit simulates projection bias and contrast reduction caused by scattering through low-frequency background and spatial correlation distribution. By combining these units, simulated projection data with various intraoperative degradation characteristics can be generated.

[0035] 3D Reconstruction Module: This module converts simulated projection data into 3D simulated medical images. Depending on the application requirements, the 3D reconstruction module can employ FDK analytical reconstruction, iterative reconstruction, or deep learning reconstruction methods. In rapid simulation scenarios, FDK analytical reconstruction can be used to improve data generation efficiency; in scenarios requiring higher quality reconstruction or stronger prior constraints, iterative reconstruction or deep learning reconstruction can be used. Through this module, degradation factors introduced in the projection domain, such as motion, beam hardening, noise, and scattering, can be represented in the reconstruction domain as intraoperative cone-beam CT features such as stripes, ghosting, blurring, intensity bias, and noise enhancement.

[0036] The structural prior generation module extracts anatomical structure-related information from basic medical images or initial simulated medical images. Structural priors can be obtained through thresholding, clustering methods, edge detection, organ segmentation networks, or deformation field estimation methods. These structural priors can represent tissue partitions such as air, soft tissue, and bone tissue, or they can represent organ regions, boundary contours, or multi-scale structural features. By incorporating structural priors into the subsequent generation network, explicit anatomical constraints can be provided for the domain enhancement process of simulated images, reducing the risk of structural drift during the generation process.

[0037] Deep Generation Enhancement Module: This module further improves the consistency of domain distribution between simulated medical images and real intraoperative cone-beam CT images. The deep generation enhancement module can employ generative adversarial networks, cyclic consistency generation networks, diffusion models, Transformer networks, Mamba networks, or combinations thereof. In one implementation, the generation network adopts a two-branch structure with an image input branch and a structural prior input branch. The image input branch is responsible for learning grayscale, texture, and artifact features, while the structural prior input branch provides anatomical constraints. A long-range dependency modeling module is introduced into the network to enhance its ability to represent global structure and high-frequency details. A discriminator is used to constrain the consistency of the generated results with the real intraoperative cone-beam CT domain distribution. During training, joint optimization can be performed using at least one of adversarial constraints, cyclic consistency constraints, identity mapping constraints, structural consistency constraints, reconstruction constraints, or perceptual constraints to ensure that the generated results are stylistically close to real intraoperative images and structurally consistent with the input base images.

[0038] Furthermore, this invention can employ a multi-stage or joint training approach to complete simulation and generative enhancement. First, a physical degradation modeling module generates initial simulated medical images with controllable degradation conditions. Then, a depth-based generative enhancement module with structural prior constraints learns the domain differences between the initial simulated images and real intraoperative cone-beam CT images, and performs style and detail enhancement on the initial simulated images. Through this approach, this invention can rapidly generate simulated medical image data with structural reference, controllable degradation, and realistic domain features, even in the absence of large-scale, rigorously paired preoperative and intraoperative data. This provides a data foundation for medical image registration, image reconstruction, cross-modal generation, intraoperative navigation, and algorithm validation.

[0039] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A rapid medical image data simulation method based on physical degradation modeling, characterized in that, First, three-dimensional medical image data is acquired and standardized preprocessed to obtain basic medical image volume data. Second, based on the scanning geometry parameters of the intraoperative cone-beam CT imaging system, the basic medical image volume data is forward-projected to obtain ideal projection data. Then, based on the ideal projection data, physical degradation factors related to intraoperative imaging are introduced, including at least one of motion artifacts, beam hardening artifacts, photon statistical noise, and scattering effects, to generate simulated projection data with intraoperative imaging degradation characteristics. Next, three-dimensional reconstruction is performed on the simulated projection data to obtain initial simulated medical image data. Subsequently, structural prior information is constructed based on the basic medical image data and the initial simulated medical image data, and the basic medical image data, the initial simulated medical image data, and the structural prior information are input into a depth generation network for intraoperative imaging domain feature enhancement. Finally, target simulated medical image data with intraoperative cone-beam CT imaging characteristics and maintaining anatomical structural consistency is output.

2. The rapid medical image data simulation method based on physical degradation modeling as described in claim 1, characterized in that, The method includes the following steps: 1) Image acquisition and preprocessing: Acquire three-dimensional medical image data; preprocess the three-dimensional medical image data, the preprocessing including at least one of resampling, cropping, intensity normalization, spatial alignment, region of interest extraction or attenuation coefficient conversion, to obtain basic medical image volume data suitable for projection simulation.

3. The rapid medical image data simulation method based on physical degradation modeling as described in claim 2, characterized in that, The method further includes the following steps: 2) Projection geometry construction: Cone-beam projection geometry model is constructed based on at least one of the following parameters of the target intraoperative cone-beam CT system: X-ray source location, detector parameters, scanning angle range, number of projections, source-to-detector distance, or source-to-rotation center distance. Based on the cone-beam projection geometry model, the basic medical image volume data is forward-projected to obtain ideal projection data, which serves as the input for subsequent physical degradation modeling.

4. The rapid medical image data simulation method based on physical degradation modeling as described in claim 3, characterized in that, The method further includes the following steps: 3) Motion artifact simulation: Introduce sampling angle perturbation, spatial position perturbation, or changes in the motion state of anatomical structures into the projection angle sequence to obtain motion projection data containing motion inconsistencies; 4) Beam hardening artifact simulation: Material regions are divided based on the gray values, tissue density, or tissue labels of basic medical imaging volume data, and material-related nonlinear attenuation mapping is applied to different material regions to obtain projection data with beam hardening characteristics. 5) Photon noise and scattering simulation: The projection data is converted to the intensity domain, and photon statistical noise and scattering components are introduced into the intensity domain. Then, the simulated projection data is obtained by normalization and logarithmic transformation.

5. The rapid medical image data simulation method based on physical degradation modeling as described in claim 4, characterized in that, The method further includes the following steps: 6) Generating initial simulated images from 3D reconstruction: Input the simulated projection data obtained by physical degradation modeling into the 3D reconstruction algorithm; perform 3D reconstruction using at least one of the following methods: FDK analytical reconstruction, algebraic iterative reconstruction, statistical iterative reconstruction, prior constraint reconstruction, or deep learning reconstruction; output initial simulated medical image data with intraoperative cone-beam CT imaging degradation characteristics.

6. The rapid medical image data simulation method based on physical degradation modeling as described in claim 5, characterized in that, The method further includes the following steps: 7) Structural Prior Construction: Structural prior information is generated based on basic medical imaging data and initial simulated medical imaging data. The structural prior information includes at least one of tissue partition mask, organ mask, edge map, anatomical label, deformation field, multi-scale structural features, or structural label map obtained by clustering method. 8) Deep Generative Network Enhancement: The basic medical image data, initial simulated medical image data, and structural prior information are used as inputs to the deep generative network; Intraoperative cone-beam CT imaging domain feature enhancement is performed on the input image using a deep generative network, so that the generated result has at least one of the following features: noise texture, scattering bias, beam hardening artifact, motion artifact, local contrast variation, or cone-beam CT image style. 9) Output: Output the target simulated medical image data.

7. The rapid medical image data simulation method based on physical degradation modeling as described in claim 4, characterized in that, In step 3), the motion artifact simulation process is as follows: based on the patient's body movement, respiratory movement, organ movement, bed movement, equipment vibration, or sampling angle error that may exist during the intraoperative scanning process, the motion disturbance type and disturbance intensity are set; some angles are selected as motion disturbance angles in the projection angle sequence, and the distribution of the motion disturbance angles is controlled within the scanning angle range; sampling angle offset is introduced into the motion disturbance angles, and spatial position offset, rotational disturbance, or non-rigid deformation is introduced into the corresponding basic medical image volume data; forward projection is performed based on the disturbed sampling angles and the disturbed basic medical image volume data to obtain motion projection data containing motion inconsistencies.

8. The rapid medical image data simulation method based on physical degradation modeling as described in claim 4, characterized in that, In step 4), the process of simulating beam hardening artifacts is as follows: based on the grayscale value, CT value, tissue density, or tissue label of the basic medical image volume data, the basic medical image volume data is divided into at least one of bone tissue region, soft tissue region, air region, or other material region; projection calculations are performed on different material regions to obtain the projection contribution corresponding to different materials, and then a material-related nonlinear attenuation mapping is applied to them; the results of different materials are synthesized to obtain projection data with beam hardening artifact characteristics.

9. The rapid medical image data simulation method based on physical degradation modeling as described in claim 4, characterized in that, In step 5), the process of simulating photon noise and scattering is as follows: the projection data is converted to the intensity domain to obtain the corresponding ray intensity data; according to the preset dose level, photon statistical noise is introduced into the ray intensity data; at least one scattering component from low-frequency scattering background, spatially correlated scattering mask, or scattering kernel response is added to the ray intensity data. Normalize and logarithmically transform the intensity data after adding photon statistical noise and scattering components to obtain simulated projection data.

10. The rapid medical image data simulation method based on physical degradation modeling as described in claim 6, characterized in that, In step 8), the deep generative network includes at least one of generative adversarial networks, recurrent consistency generative networks, diffusion models, Transformer networks, Mamba networks, selective state-space model networks, convolutional neural networks, or combinations thereof. The deep generative network includes an image input branch and a structural prior input branch, wherein the image input branch is used to extract grayscale, texture, and artifact features, and the structural prior input branch is used to provide anatomical structural constraints. The deep generative network includes a long-range dependency modeling module, which is used to enhance the expressive power of global structure, edge details, texture distribution, or artifact features. The deep generative network is trained by at least one of adversarial constraints, recurrent consistency constraints, identity mapping constraints, structural consistency constraints, reconstruction constraints, or perceptual constraints.

Images