Cardiac coronary plaque simulation method and system based on digital phantom

Abstract

The invention belongs to the technical field of medical image data processing, and particularly relates to a cardiac coronary plaque simulation method and system based on a digital phantom, and the method comprises the step of generating reconstructed simulation CT projection images with different halo artifact degrees. The method solves the problems that the effectiveness of an artifact removal method cannot be verified and evaluated due to the fact that the algorithm training effect is not generalized enough, and has the beneficial technical effects that the effectiveness of the artifact removal method is fully verified and evaluated, and the generalization of a deep learning halo artifact removal algorithm is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of medical image data processing, and particularly relates to a coronary plaque simulation method and system based on a digital phantom. BACKGROUND

[0002] Coronary computed tomography angiography (CTA) can directly assess coronary stenosis, which is the first-line examination for acute and chronic coronary syndromes; the coronary calcified plaque segmentation module plays an important role in precise diagnosis and treatment, but still faces multiple challenges such as lack of data and low quality, algorithm generalization, etc.; the main challenge of coronary CTA is the image quality of calcified plaque lesions, and the diagnostic accuracy is affected by the halo artifact of calcified plaque; due to the halo artifact caused by calcified plaque, the CT value (Hounsfield Unit, HU) of the adjacent tissue deviates, thereby masking the adjacent tissue and falsely increasing the stenosis area; currently, various solutions have been proposed to solve the problem of halo artifact, including protocol optimization of image acquisition, hardware improvement and post-processing algorithm optimization; usually, comparative tests are carried out with other imaging modalities (such as optical coherence tomography) or physical phantoms to verify and evaluate the effectiveness of the de-artifact method;

[0003] The geometry of the physical phantom is usually regular, and the material composition is limited, which faces challenges in simulating complex anatomical structures; in addition, the material of the physical phantom may degrade over time, resulting in inaccurate results; the digital phantom can simulate the effect of CT scanning protocol on calcified plaque halo artifact, and promote related research on removing halo artifact by simulating various clinical situations; the simulated images obtained from the digital phantom can be used for algorithm research and regulatory evidence research; in contrast, the digital phantom can quickly and accurately generate various anatomical shapes and materials, realize the simulation of complex anatomical structures, and can be expanded by adding surrounding tissues (such as myocardium, bone), thereby realizing fast and low-cost evaluation, providing more objective and clinically relevant evaluation; in addition, the true structure of the digital phantom is accurate and known, which makes it more advantageous for comparison and verification of plaque segmentation methods;

[0004] In summary, the prior art has the problems of being unable to verify and evaluate the effectiveness of the de-artifact method due to the single and simple sample provided by the sample, and the generalization of the algorithm training effect is insufficient. SUMMARY

[0005] The present application provides a coronary plaque simulation method and system based on a digital phantom to solve the above problems in the background art. Specifically, the technical problem solved by the present application is realized by the following technical solution:

[0006] The simulation method of a heart coronary artery plaque based on a digital phantom comprises the following steps:

[0007] Based on clinical heart coronary artery data, a heart coronary artery basic model with different lesion degrees is generated through automatic region labeling or manual modeling, wherein the different lesion degrees include different stenosis degrees, different plaque shapes and different plaque material degrees.

[0008] Based on the heart coronary artery basic model, a digital heart coronary artery plaque phantom with different material tissues is generated through material parameter setting, wherein the different material tissues include a coronary artery vessel, a contrast agent, a calcified plaque, fibrous tissue and lipid tissue.

[0009] Based on the digital heart coronary artery plaque phantom, a simulation CT projection image is generated through simulation CT imaging according to simulation environment setting, wherein the simulation environment setting includes radiation source simulation setting, tube voltage and its commonly used range simulation setting, photon emission number simulation setting of each projection angle, detector pixel size setting and detector imaging matrix setting.

[0010] Based on the simulation CT projection image, a reconstructed simulation CT projection image with different halo artifact degrees is generated through a reconstruction filter kernel, wherein the reconstruction filter kernel includes a smooth filter kernel and a sharp filter kernel.

[0011] Based on the reconstructed simulation CT projection image, an algorithm deartifact result is obtained by introducing it into a current halo artifact removal algorithm, and the algorithm deartifact result is compared with different virtual lesion setting errors to determine the accuracy of the current halo artifact removal algorithm.

[0012] Preferably, the automatic region labeling comprises:

[0013] Based on the clinical heart coronary artery data, denoised vessel data is generated through median filter denoising.

[0014] Based on the denoised vessel data, vessel edge features are generated through vessel edge fitting.

[0015] Based on the vessel edge features, a high-contrast vessel image is generated through histogram equalization enhancement.

[0016] Based on the high-contrast vessel image, a high-contrast vessel grayscale image is generated through image weighted average.

[0017] Based on the high-contrast vessel grayscale image, a labeled vessel image is generated according to a target labeling rule based on image CT value.

[0018] Based on the labeled blood vessel image, a binary blood vessel image with a determined labeled region is generated through binary processing, wherein the determined labeled region includes a heart, a coronary artery blood vessel and a calcified plaque.

[0019] Based on the binary blood vessel image with the determined labeled region, a three-dimensional model file corresponding to a digital heart coronary artery plaque phantom is generated through labeled region extraction.

[0020] Preferably, the target labeling rule based on the image CT value includes:

[0021] If the static image CT value of the current region is 30-50HU, the current region is marked as myocardial tissue;

[0022] If the static image CT value of the current region is 40-60HU, the current region is initially judged as a blood vessel tissue, and if the current region is initially judged as a blood vessel tissue, if the blood flow image CT value of the current region is 30-50HU, the current region is marked as a coronary artery blood vessel;

[0023] If the static image CT value of the current region is ≥130HU and the image edge is sharp and shows high-density spots or sheets, the current region is initially judged as a typical calcified plaque, and if the current region is initially judged as a typical calcified plaque, further judgment is made:

[0024] If the static image CT value of the current region is 130-199HU, the current region is marked as a mild calcified plaque;

[0025] If the static image CT value of the current region is 200-299HU, the current region is marked as a moderate calcified plaque;

[0026] If the static image CT value of the current region is ≥300HU, the current region is marked as a severe calcified plaque.

[0027] Preferably, the simulation CT imaging includes:

[0028] Based on the digital heart coronary artery plaque phantom data, preprocessed phantom data is generated through phantom data generation;

[0029] Based on the radiation source basic parameters, the tube voltage basic parameters and the photon emission basic parameters, a simulation environment setting is generated through simulation parameter setting;

[0030] Based on the preprocessed phantom data and the simulation environment setting, a simulation CT projection image is generated through virtual rotating scanning.

[0031] Preferably, the phantom data generation includes:

[0032] Based on the digital heart coronary plaque phantom data, a digital phantom metadata set is generated through bio-information analysis;

[0033] Based on the digital phantom metadata set, a low-noise phantom data set is generated through low-quality data filtering;

[0034] Based on the low-noise phantom data set, a standard phantom data set with different labeled regions is generated through optimization of display contrast, wherein the different labeled regions include heart, coronary blood vessels, and calcified plaques;

[0035] The simulation parameter setting includes:

[0036] Based on the radiation source basic parameters, a radiation source simulation setting is generated through radiation source simulation parameter setting;

[0037] Based on the tube voltage basic parameters, a tube voltage and its commonly used range simulation setting is generated through tube voltage simulation parameter setting;

[0038] Based on the photon emission basic parameters, a photon emission number simulation setting for the current projection angle is generated through photon emission simulation parameter setting;

[0039] Based on the radiation source simulation setting, the tube voltage and its commonly used range simulation setting, and the photon emission number simulation setting for each projection angle, the simulation parameter setting is aggregated;

[0040] The virtual rotational scan includes:

[0041] Based on the preset angle sequence, the detector is rotated to the preset angle through the virtual transmission pulse signal, wherein the preset angle sequence includes, for example, 0°, 180°, 360°, 0°;

[0042] Based on the preprocessed phantom data, the detector is rotated and scanned through the spiral CT virtual mode according to the simulation environment setting, wherein the spiral CT virtual mode includes rotating with a step size ≤0.1 and covering 360° rotation continuously;

[0043] When the detector is rotated to the preset angle and stabilized, virtual X-ray exposure is performed and the detector data acquisition thread is started to generate a virtual exposure image at the current angle;

[0044] Based on the virtual exposure image at the current angle, a multi-angle simulation CT projection image set containing angle labels, time stamps, and other metadata is generated through exposure image combination.

[0045] Preferably, the reconstruction filter kernel includes:

[0046] Based on the simulation CT projection image, a reinforced CT image is generated through weighted processing;

[0047] Based on the reinforced CT image, according to the smooth filter kernel function and the sharp filter kernel function processing, the reconstructed simulation CT projection image with different halo artifact degrees is generated by adjusting the coefficient and the cut-off frequency optimization.

[0048] Preferably, the smooth filter kernel function:

[0049]

[0050] Wherein, the H'(f) is the frequency response after the smooth filter kernel processing, the f m is the cut-off frequency, the f is the original frequency;

[0051] The sharp filter kernel function:

[0052]

[0053] The H''(f) is the frequency response after the sharp filter kernel processing, the a is the first weighting coefficient, and the b is the second weighting coefficient.

[0054] Preferably, the reconstruction filter kernel further includes a visualization optimization, and the visualization optimization includes:

[0055] The visualization optimization function:

[0056]

[0057] The mu is a linear attenuation coefficient, the u_water is an attenuation value of water, and the u_object is an attenuation value of an object, wherein the object is a digital heart coronary plaque phantom, and the CT attenuation value is the simulation CT data after the visualization optimization processing.

[0058] Preferably, the simulation method further includes:

[0059] Based on the simulation CT projection image of different virtual lesions, the simulation CT sample is generated by data set sample construction, and finally used as a training sample for current halo artifact training.

[0060] Meanwhile, the application also provides a heart coronary plaque simulation system based on a digital phantom, including:

[0061] Based on the simulation method according to any one of the above, the simulation system includes a simulation platform, and the simulation platform includes:

[0062] The module for generating a basic model of a heart coronary artery comprises the following steps: based on clinical heart coronary artery data, a basic model of a heart coronary artery with different lesion degrees is generated through automatic region labeling or manual modeling, wherein the different lesion degrees include different stenosis degrees, different plaque shapes and different plaque material degrees.

[0063] The module for generating a digital heart coronary artery plaque phantom comprises the following steps: based on the basic model of the heart coronary artery, a digital heart coronary artery plaque phantom with different material tissues is generated through material parameter setting, wherein the different material tissues include a coronary artery vessel, a contrast agent, a calcified plaque, fibrous tissue and lipid tissue.

[0064] The module for generating a simulation CT projection image comprises the following steps: based on the digital heart coronary artery plaque phantom, a simulation CT projection image is generated through simulation CT imaging according to simulation environment setting, wherein the simulation environment setting includes radiation source simulation setting, tube voltage and commonly used range simulation setting, photon emission number simulation setting of each projection angle, detector pixel size setting and detector imaging matrix setting.

[0065] The module for generating a reconstructed simulation CT projection image comprises the following steps: based on the simulation CT projection image, a reconstructed simulation CT projection image with different halo artifact degrees is generated through a reconstruction filter kernel, wherein the reconstruction filter kernel includes a smooth filter kernel and a sharp filter kernel.

[0066] The module for judging the accuracy of a current halo artifact removal algorithm comprises the following steps: based on the reconstructed simulation CT projection image, the algorithm deartifact result is obtained by introducing the reconstructed simulation CT projection image into the current halo artifact removal algorithm, the algorithm deartifact result is compared with different virtual lesion setting errors, and then the accuracy of the current halo artifact removal algorithm is judged.

[0067] Beneficial technical effects:

[0068] The present scheme generally comprises five steps, namely, the first step of generating a basic model of a heart coronary artery with different lesion degrees, the second step of generating a digital heart coronary artery plaque phantom with different material tissues, the third step of generating a simulation CT projection image, the fourth step of generating a reconstructed simulation CT projection image with different halo artifact degrees, and the fifth step of obtaining an algorithm deartifact result or a training sample for current halo artifact removal training. Through the above steps, a digital blood vessel model with different plaque geometric shapes and stenosis degrees is developed, and simulation CT imaging is realized. By adjusting the tube voltage and the reconstruction kernel in the simulation imaging process, it is found that these two factors significantly affect the degree of calcified halo artifact. Through the more complex and diverse characteristics of the digital phantom model, the effectiveness of the deartifact method is fully verified and evaluated. At the same time, by more accurately measuring and simulating various clinical conditions, halo artifact is generated by adjusting the imaging parameters to generate various data sets, and the generalization of the deep learning halo artifact removal algorithm is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Fig. 1 is a flow chart of a method for simulating coronary artery plaque of the present invention;

[0070] Fig. 2 This is an application flow chart of the coronary artery plaque simulation method of the present invention;

[0071] Fig. 3 This is a key structural diagram of the digital phantom of the coronary artery plaque simulation method of the present invention;

[0072] Fig. 4 It is a schematic diagram of the process of generating a simulated CT image of the coronary artery plaque simulation method of the present invention;

[0073] Fig. 5 This is an illustration of calcified materials with different CT values ​​in the coronary artery plaque simulation method of the present invention;

[0074] Fig. 6 Schematic diagram of different window functions of the reconstruction filter of the coronary artery plaque simulation method of the present invention. DETAILED DESCRIPTION

[0075] The present invention will be further described below with reference to the accompanying drawings:

[0076] In the picture:

[0077] S101 - Generate basic models of coronary arteries with different degrees of lesions; S1001 - Generate basic models of coronary arteries with different degrees of lesions based on clinical coronary artery data through automatic region annotation or manual modeling;

[0078] S102 - Generate digital coronary artery plaque phantoms of different materials and tissues; S1002 - Generate digital coronary artery plaque phantoms of different materials and tissues based on the basic coronary artery model by setting material parameters;

[0079] S103 - generating a simulated CT projection image; S1003 - generating a simulated CT projection image through simulated CT imaging based on the digital heart coronary artery plaque phantom and according to the simulation environment settings;

[0080] S104 - generating reconstructed simulated CT projection images with different halo artifact degrees; S1004 - generating reconstructed simulated CT projection images with different halo artifact degrees based on the simulated CT projection images by reconstructing the filter kernel;

[0081] S105 - get the algorithm de-pseudo result; S1005 - based on the reconstructed simulation CT projection image, by introducing it into the current de-halo pseudo algorithm, get the algorithm de-pseudo result, judge the algorithm de-pseudo result and the error of different virtual lesions, and further judge the accuracy of the current de-halo pseudo algorithm;

[0082] S106 - as the training sample of the current de-halo pseudo training; S1006 - based on the simulation CT image of different virtual lesions, the simulation CT sample is generated through the data set sample construction, and finally it is used as the training sample of the current de-halo pseudo training; Embodiment:

[0083] As shown in Figs. 1-2 The heart coronary artery plaque simulation method based on digital phantom comprises:

[0084] Based on the clinical heart coronary artery data, the heart coronary artery basic model with different lesion degrees is generated through automatic region annotation or manual modeling S101 S1001, wherein the different lesion degrees include different stenosis degrees, different plaque shapes and different plaque material lesion degrees;

[0085] Based on the heart coronary artery basic model, the digital heart coronary artery plaque phantom of different material tissues is generated through material parameter setting S102 S1002, and the different material tissues include coronary artery, contrast agent, calcified plaque, fibrous tissue and lipid tissue;

[0086] Based on the digital heart coronary artery plaque phantom, the simulation CT projection image is generated through simulation CT imaging according to the simulation environment setting S103 S1003; wherein the simulation environment setting includes radiation source simulation setting, tube voltage and its commonly used range simulation setting, photon emission number simulation setting of each projection angle, detector pixel size setting and detector imaging matrix setting;

[0087] Based on the simulation CT projection image, the reconstructed simulation CT projection image with different halo pseudo image degrees is generated through the reconstruction filter kernel S104 S1004, wherein the reconstruction filter kernel includes the construction of smooth filter kernel and sharp filter kernel;

[0088] Based on the reconstructed simulation CT projection image, by introducing it into the current de-halo pseudo algorithm, the algorithm de-pseudo result is obtained S105, the algorithm de-pseudo result and the error of different virtual lesions are judged, and further the accuracy of the current de-halo pseudo algorithm is judged S1005.

[0089] The present application comprises five steps, namely: the first step: generating digital phantoms of different lesion degrees; the second step: generating simulated CT images; the third step: generating reconstructed simulated CT images of different halo artifact degrees; the fourth step: generating simulated CT images of different virtual lesion settings; and the fifth step: obtaining algorithm de-artifact results or serving as training samples for current halo artifact removal training.

[0090] The specific scheme of the present application is: the present application comprises a digital phantom design module, a CT simulation imaging module and a CT projection image reconstruction module. The digital phantom design module is used to design a model of coronary arteries and plaques and set material parameters of different plaques. The CT simulation imaging module is used to set relevant parameters of a CT simulation imaging system, including parameters of a radiation source and a detector. The CT projection image reconstruction module is used to reconstruct CT projection images obtained through simulation imaging, and by adjusting the smoothing and sharpness of a reconstruction filter kernel, simulated coronary CTA images of different halo artifact degrees can be obtained. The present application is used for auxiliary analysis of halo artifacts of calcified plaques of heart coronary arteries, and by modifying relevant parameters, simulated coronary CTA images of different halo artifact degrees can be obtained. The present application is used for auxiliary analysis of halo artifacts of calcified plaques of heart coronary arteries, and by using coronary CTA images, it is of great significance to the diagnosis of halo artifacts of calcified plaques of heart coronary arteries and the optimization and improvement of de-artifact algorithms. Specifically:

[0091] Step 1: constructing a digital phantom (a model design unit of coronary arteries and plaques and a plaque material parameter setting unit)

[0092] Based on a three-dimensional modeling tool, coronary artery models of different stenosis degrees (10% to 90% and the like) and different plaque shapes and materials (fibers, lipids and calcification) are constructed. Step 2: simulation CT imaging, based on a Monte Carlo simulation platform (GATE), parameters of simulation CT (including: a radiation source (photon number, voltage), a detector (pixel size, total number), body model material parameters (fibers, lipids, calcification) and the like) are defined, and simulation CT imaging is performed. Step 3: simulation CT projection image reconstruction, projection images obtained through a simulation CT platform are reconstructed, and a reconstruction filter kernel (smoothing, sharpness) is designed, and simulation CT images are obtained. Step 4: visualization results, in the present application, the gray scale values of the reconstructed images reflect the linear attenuation coefficients (μ) of the materials, which can be converted into CT values (unit: HU) through a formula. Step 5: algorithm verification or training based on simulation CT images, by using the simulation CT images of different halo artifact degrees obtained, the halo artifact removal algorithm is verified; or by using the simulation CT images, a data set is constructed, and the AI-based halo artifact removal algorithm is trained.

[0093] Preferably, the present application develops a new method to simulate the CT image of the coronary artery by a series of digital blood vessel phantoms to generate halo artifacts of calcified plaques. The model can be used to study the influence of X-ray tube voltage and reconstruction kernel on the severity of halo artifacts of calcified plaques. The innovation points are: (1) the definition of materials (fiber, lipid, contrast agent, calcified plaque) corresponds to different material CT values; (2) the shape design of the plaque corresponds to the shape of the calcified / lipid / fiber plaque of different shapes; (3) the innovation of the calculation modeling method: a multi-parameter model (combination of reconstruction filter kernel, tube voltage sharpness, detector size) is proposed to generate simulated halo artifacts of calcified plaques, which corresponds to different sizes of halo artifacts of calcified plaques, and solves the problem of large change in artifact shape in traditional CT simulation. (4) Cross-platform verification method: for the first time, 3D printing physical models are combined with CT equipment from multiple manufacturers to verify and establish a standardized evaluation system for digital phantoms. (5) Through computer modeling and simulation, a large amount of data is generated as special data for AI training.

[0094] The present application develops digital blood vessel models with different plaque geometries and stenosis severity and realizes simulated CT imaging through the above steps; by adjusting the tube voltage and reconstruction kernel during the simulation imaging process, it is found that these two factors significantly affect the degree of calcified halo artifacts, and through the more complex and diverse characteristics of the digital phantom model, the effectiveness of the de-artifact method is fully verified and evaluated, at the same time, through more accurate measurement and simulation of various clinical situations, by adjusting the imaging parameters to produce halo artifacts to produce various data sets, the generalization of the deep learning de-halo artifact algorithm is improved.

[0095] As shown in Fig. 3 , 5 , the automatic region labeling includes:

[0096] Based on clinical heart coronary artery data, denoising blood vessel data is generated through median filtering denoising;

[0097] Based on the denoised blood vessel data, blood vessel edge features are generated through blood vessel edge fitting;

[0098] Based on the blood vessel edge features, high-contrast blood vessel images are generated through histogram equalization enhancement;

[0099] Based on the high-contrast blood vessel images, high-contrast blood vessel grayscale images are generated through image weighted average;

[0100] Based on the high-contrast blood vessel grayscale images, labeled blood vessel images are generated according to the target labeling rules based on image CT values;

[0101] Based on the labeled blood vessel image, a binary blood vessel image with a determined labeled region is generated through binarization processing, wherein the determined labeled region includes a heart, a coronary artery blood vessel, and a calcified plaque.

[0102] Based on the binary blood vessel image with the determined labeled region, a three-dimensional model file corresponding to a digital heart coronary artery plaque phantom is generated through labeled region extraction.

[0103] Regarding the construction of coronary artery models with different stenosis degrees such as 10% to 90%, two different methods can be used, namely automatic region labeling or manual modeling. Regarding automatic region labeling, the target regions of clinical data are labeled (such as heart, coronary artery blood vessel, and calcified plaque), the labeled regions are extracted and derived to obtain a three-dimensional model file (such as STL, OBJ, etc.), and different target regions have different CT values. The target regions of clinical data are labeled based on CT values (such as heart, coronary artery blood vessel, and calcified plaque), and similarly, different materials such as fiber, lipid, and calcification can be distinguished. The specific method is as follows: eliminate salt and pepper noise through median filtering and retain blood vessel edge features; enhance the contrast between blood vessels and background through histogram equalization to improve the visibility of weak blood vessel regions; convert the color image to a grayscale image, usually using the weighted average method (such as weighted average RGB: R: 0.3, G: 0.59, B: 0.11); fixed threshold: based on prior knowledge or histogram bimodal characteristics, manually set a fixed threshold to convert the grayscale image to a binary image, thereby determining the heart, coronary artery blood vessel, and calcified plaque region.

[0104] In addition, regarding manual modeling, a three-dimensional modeling tool (such as UG NX, SolidWorks, etc.) is used to model the target region (such as heart, coronary artery blood vessel, and calcified plaque) to obtain a three-dimensional model file (such as STL, OBJ, etc.). Fig. 5 When simulating CT, the material definition of the simulation object (such as the coronary plaque model) directly affects the physical process of photon interaction (such as attenuation and scattering), and then determines the authenticity of the imaging result. The parameters of material definition are as follows: Material Name (Material Name): a self-defined identifier (such as Water, Bone, Tungsten); Density (Density): unit g / cm 3 , directly affecting the attenuation coefficient (such as the density of water is 1.0 g / cm 3 , the density of bone is 1.5-2.0 g / cm 3) ; Element Composition composite / mixture defines components by fractional mass or number of atoms; in the present application, based on CT simulation tool, the composition elements of the material and the proportion of each element, the density of the material are defined to set the material parameters, see the table below. By adjusting the definition of the material (especially calcified plaque) (such as composition elements and proportion), different material CT values can be obtained; the innovation: for the definition of different materials (fibers, lipids, contrast agents, calcified plaque), and adjusting the composition element and proportion parameters, the density value, after simulation, the corresponding different material CT values can be obtained, especially the calcified plaque at different development stages (by adjusting the weighting coefficient, the calcified material with different CT values is obtained).

[0105] The target labeling rule based on the image CT value comprises:

[0106] If the static image CT value of the current region is 30-50HU, the current region is marked as myocardial tissue;

[0107] If the static image CT value of the current region is 40-60HU, the current region is initially judged as blood vessel tissue, and if the current region is initially judged as blood vessel tissue, if the blood flow image CT value of the current region is 30-50HU, the current region is marked as coronary artery blood vessel;

[0108] If the static image CT value of the current region is ≥130HU and the image edge is sharp and shows high-density spot or sheet, the current region is initially judged as typical calcified plaque, and if the current region is initially judged as typical calcified plaque, further judgment is made:

[0109] If the static image CT value of the current region is 130-199HU, the current region is marked as mild calcified plaque;

[0110] If the static image CT value of the current region is 200-299HU, the current region is marked as moderate calcified plaque;

[0111] If the static image CT value of the current region is ≥300HU, the current region is marked as severe calcified plaque.

[0112] The fixed threshold of the application is based on prior knowledge or histogram bimodal characteristics, and the fixed threshold is manually set. Regarding the fixed threshold: the myocardial tissue fixed threshold is generally 30-50 HU, which belongs to soft tissue density, and its image characteristics are uniform moderate density, and after enhanced scanning, it can be slightly elevated due to blood perfusion, but it is lower than the density of the intravascular contrast agent; the unenhanced CT fixed threshold: the density of the blood vessel wall is close to the surrounding soft tissue (about 40-60 HU), and the lumen is low in density (about 30-50 HU) due to blood flow, and the enhanced CTA fixed threshold (including contrast agent): the lumen is filled with iodine contrast agent, and the CT value is significantly increased to 200-600 HU, and the boundary is clear; the typical calcified plaque fixed threshold CT threshold: ≥130 HU, the edge is sharp, and it is a high-density spot or sheet; the classification basis: 130-199 HU: mild calcification; 200-299 HU: moderate calcification; ≥300 HU: severe calcification, and the calcified plaque is located along the distribution of the coronary artery (such as the proximal segment of the left anterior descending branch and the middle segment of the right coronary artery), and needs to be combined with multi-planar reconstruction (MPR) to confirm;

[0113] Computer modeling and simulation have gradually become a method for medical device product validation. The application designs more specialized blood vessel calculation models for optimizing or validating coronary CTA imaging and analysis. Compared with actual CT scanning, the image quality obtained by simulation CT imaging is more stable. In clinical practice, inconsistent calcified plaque classification and scanning schemes may affect the quantitative analysis results and introduce inter-observer variability. By avoiding modeling errors (including material performance degradation, inaccurate model formulation) and acquisition errors (such as aberration, incorrect positioning and motion artifacts), the method proposed in the application can achieve more consistent measurements and reduce model variability; the application designs 20 calculation models (STL format), simulates blood vessel segments with different degrees of stenosis, including coronary blood vessel wall, lumen, fibrous tissue, lipid plaque and calcified plaque, and the geometric shape of each model is consistent along the axial direction. The length, wall thickness and cross-sectional outer diameter of the blood vessel model are 15, 1 and 8 mm respectively. In order to represent the variability of the plaque, five models are defined, named Model 1-5. The total area of the plaque (including fibrous tissue, lipid plaque and calcified plaque) is set to 10.33 mm2, 15.75 mm2, 19.40 mm2 and 23.82 mm2, corresponding to four degrees of stenosis (mild, moderate, advanced and severe), and the area stenosis percentage is 36.54%, 55.71%, 68.62% and 84.26%. The area of the calcified plaque ranges from 4.95 to 23.82 mm2.

[0114] The composition of calcified plaque is defined as dense calcified plaque with a CT value ranging from 500 to 1200 HU (under the condition of a collection voltage of 100 kV). Based on the material definition of fat and muscle in GateMaterials.db (the material database of GATE), the material definition of fibrous tissue and lipid plaque is modified to match the CT values of fibrous tissue (131-350 HU) and lipid plaque (-42-47 HU). Based on the molecular formula of iodine billi do (XENETIX 350), the composition of the contrast agent is defined.

[0115] As shown in Fig. 4 The simulation CT imaging includes:

[0116] Based on the digital heart coronary plaque phantom data, the preprocessed phantom data is generated through phantom data generation;

[0117] Based on the basic parameters of the radiation source, the basic parameters of the tube voltage, and the basic parameters of the photon emission, the simulation environment setting is generated through simulation parameter setting;

[0118] Based on the preprocessed phantom data and the simulation environment setting, the simulation CT projection image S103 is generated through virtual rotating scanning.

[0119] The way of generating the simulation CT image is to simulate the actual clinical imaging environment by specifying the conditions of the simulation CT, refer to the protocol of CT heart coronary imaging, and obtain the simulated CT image. At the same time, in order to improve the simulation operation efficiency, parallel operation is carried out based on the supercomputing center;

[0120] Radiation source setting: during simulation, the definition of the radiation source is one of the key steps, and the parameter setting directly affects the accuracy and authenticity of the simulation. The radiation source type setting: the optional range of the radiation source type includes point source, line source, and surface source; in the present application, the preferred value is point source.

[0121] Tube voltage value and common range selection: Energy parameter setting and tube voltage (kVp) definition: In CT simulation, the tube voltage (kVp) of the X-ray tube determines the maximum energy and energy spectrum shape of the photons, directly affecting the imaging contrast and dose distribution. Usually, the tube voltage is set by using a simulated continuous energy spectrum (real model). The energy parameter optional range includes single energy (fixed energy, such as 140 keV), continuous energy spectrum (defined by energy spectrum file, simulate the bremsstrahlung of the X-ray tube), energy range; Tube voltage value (such as 80 kVp, 100 kVp, 120 kVp) directly affects the penetration ability and energy spectrum distribution of X-rays; By setting the energy spectrum parameters of the X-ray source, such as setting the tube voltage (kVp) value (such as 120 kVp), and selecting the filter material (such as aluminum or copper filter) to simulate the filter sheet (such as aluminum, copper) used in actual CT, to simulate the energy spectrum hardening effect in the actual CT system; In the present application, the energy spectrum generation method 1 is to use a predefined energy spectrum file: generate an energy spectrum file (such as spectrum_120kVp.txt) from experimental measurement or simulation tool (such as Spektr, SpekCalc, IPEM78 model), and then load it into the simulation. Method 2 is to use the built-in energy spectrum generator of the simulation tool to define the energy spectrum. Tube voltage (kVp) refers to the maximum accelerating voltage of the X-ray tube (such as 80 kVp, 120 kVp), which determines the upper limit of the photon energy (such as 120 kVp→ maximum energy 120 keV). The actual energy spectrum refers to the X-ray as a continuous energy spectrum (bremsstrahlung) superimposed with characteristic peaks, simulated by an energy spectrum file. Continuous energy spectrum is more time-consuming than single energy simulation, which can be optimized by limiting the energy range (such as 20-120 keV). The halo artifact of calcified plaque is related to many factors, and the present application mainly focuses on tube voltage and reconstruction kernel. The X-ray energy spectrum is calculated, and the beam voltage is set to 80 / 100 / 120 kV. Considering the balance between image quality and simulation operation time, the preferred photon number setting for simulation running is 6x108.

[0122] Set the number of photon emissions per projection angle: Current / Photon number setting: When simulating CT, controlling the number of photon emissions per projection angle is a critical step that directly affects the statistical noise of the projection data and the computational efficiency. CT simulation usually requires rotating the X-ray source (or detector) to obtain projections at different angles. The number of photon emissions per angle is determined by parameters such as the total number of photons (user-defined), angle interval (usually 0° to 360°), etc. The number of photons needs to be balanced according to the noise level requirements and computing resources (e.g., low-dose simulation: 10^4-10^5 photons / angle; high-precision simulation: 10^6-10^8 photons / angle). In CT simulation, set the continuous energy spectrum or single-energy gamma rays (Gamma / X-ray) generated by the X-ray tube, and the number of photon emissions per projection angle (e.g., 10^8 photons / angle). The higher the number, the better the signal-to-noise ratio of the simulation results, but the longer the time-consuming calculation. Verify the image quality of the simulation through small-scale simulation (such as 10^5 photons), and gradually increase the number of photons until the details of the key areas (such as calcified plaques) are stable. If you need to simulate a low-dose CT scenario, you can reduce the number of photons (such as 10^5 photons / angle), and gradually increase the number of photons and compare the stability of the results to optimize the parameters. By setting the scanning angle range (such as 0°~360°) and step angle (such as 1°), ensure that the sampling theorem requirements of CT reconstruction are met. Combined with the quantum efficiency of the detector (such as the CZT detector efficiency of about 90%), calculate the total number of photons that meet the signal-to-noise ratio (SNR≥30dB).

[0123] Detector settings:

[0124] When simulating CT, the parameter definition of the detector directly affects the imaging quality, signal acquisition efficiency, and accuracy of the simulation results. Through parameter setting, the spatial resolution, noise characteristics, and signal response of the real CT detector can be accurately simulated, providing a reliable foundation for simulation imaging.

[0125] 1. Set the detector pixel size: from 0.2mm×0.2mm to 0.625mm×0.625mm

[0126] 2. Set the detector array size: The key parameters of the detector array size include shape, size, and pixel number; the preferred value is 512×512, or 256×256.

[0127] 3. Set the detector pixel pitch: for example, 1mm×1mm.

[0128] 4. Set the material properties: The key parameters of the material properties include the detector material, and in the present application, the preferred value is a crystal detector, such as CsI (cesium iodide), which is used to convert photons into visible light.

[0129] The phantom data generation includes:

[0130] Based on the digital heart coronary artery plaque phantom data, a digital phantom element dataset is generated through bioinformatics analysis;

[0131] Based on the digital phantom dataset, a low-noise phantom dataset is generated by filtering low-quality data;

[0132] Based on the low-noise phantom dataset, a standard phantom dataset with different annotated regions is generated by optimizing display contrast, wherein the different annotated regions include the heart, coronary arteries, and calcified plaques;

[0133] The simulation parameter settings include:

[0134] Based on the basic parameters of the radioactive source, the radioactive source simulation settings are generated through the radioactive source simulation parameter settings;

[0135] Based on the basic parameters of tube voltage, the simulation settings of tube voltage and its common range are generated through the tube voltage simulation parameter settings;

[0136] Based on the basic parameters of photon emission, the simulation settings of the number of photon emissions at the current projection angle are generated through the photon emission simulation parameter settings;

[0137] Based on the simulation settings of the radiation source, the tube voltage and its common range, and the number of photon emissions per projection angle, they are aggregated into simulation parameter settings;

[0138] The virtual rotation scan includes:

[0139] Based on a preset angle sequence, the detector is rotated to a preset angle by virtually sending a pulse signal, wherein the preset angle sequence includes, for example, 0°, 180°, 360°, and 0°;

[0140] Based on the pre-processed phantom data and according to the simulation environment settings, the detector is rotated and scanned in a spiral CT virtual mode, wherein the spiral CT virtual mode includes continuous rotation with a step size of ≤0.1 and a 360° rotation coverage;

[0141] When the detector rotates to the preset angle and stabilizes, a virtual X-ray exposure is performed and the detector data acquisition thread is started to generate a virtual exposure image at the current angle;

[0142] Based on the virtual exposure image of the current angle, a multi-angle simulated CT projection image set containing metadata such as angle tags and timestamps is generated by combining exposure images.

[0143] Post-import preprocessing: Support DICOM format batch import, automatically parse patient information, scan parameters (slice thickness, scan matrix) and other metadata. Filter or label low-quality data (motion artifacts, excessive noise). Apply window width and window level standardization (e.g. calcification area: 500-1000HU, soft tissue: 350-400HU) to optimize display contrast. Use Non-Local Means or deep learning models (e.g. U-Net) to reduce noise interference. 1. Prepare the STL file; ensure that the STL file is in the correct format and that the geometry meets the simulation requirements. STL files are commonly used to describe the surface geometry of three-dimensional objects. 2. Import the STL file in the macro file, use the command to import the STL file into the simulation environment. 3 Set other simulation parameters, in addition to importing the STL file, you also need to set other simulation parameters such as CT geometry, physical process, detector model, source definition, etc. When simulating CT, recording the photon energy distribution after penetrating the object requires the use of a SensitiveDetector. The SensitiveDetector is used to record the photon information (such as energy, position, etc.) that passes through the object. By performing a 360° continuous rotation scan, projection data is collected at each angle point (including angle labels, timestamps, etc. metadata). After running the simulation, the output is in ROOT or ASCII format. Use ROOT (CERN's data analysis tool) or Python (uproot / PyROOT) to read the data.

[0144] The specific steps include:

[0145] Step 1: Define the basic parameters of the radiation source;

[0146] Step 2: Set the total number of photons and the angle range (total number of photons, angle range (usually 0° to 360°)); determine the minimum photon density (such as at least 1000 photons per pixel) according to the imaging resolution requirements to prevent statistical noise interference. Combine the detector quantum efficiency (such as the CZT detector efficiency is about 90%) to calculate the total number of photons that meet the signal-to-noise ratio (SNR≥30dB). Verify signal convergence through small-scale simulation (such as 10^5 photons), and gradually multiply the number of photons until the details of the key area (such as calcification plaque) are stable; in spiral CT mode, cover 360° continuous rotation, step ≤0.1° to suppress motion artifacts; determine the effective receiving angle range according to the detector size (such as 144 rows of 0.4mm slice thickness4) and fan angle (such as 60°); scatter angle correction: for high atomic number materials (such as calcification plaque), set additional angle sampling points (such as ±30° to enhance sampling;

[0147] Step 3: Control the angle and photon emission through a loop;

[0148] The pulse signal is sent according to a preset angle sequence (such as 0°→180°→360°→0°), and the angle displacement is realized by counting the steps of the counter; the X-ray exposure (such as exposure time 10 ms) is triggered after each angle is stabilized, and the detector data acquisition thread is started at the same time; the 360° continuous rotation scanning is performed, the projection data at each angle point is collected and stored in the RAW format (including angle label, timestamp and other metadata); the angle coding value is added to each frame of projection data, and the timestamp is used to align the scanning frame position and the detector data stream; in the simulation, random jitter (such as ±0.5° disturbance) artifacts are added, and the control stability is verified through adjacent frame correlation analysis;

[0149] As shown in Fig. 6 , the reconstruction filter core includes:

[0150] Based on the simulation CT image, a reinforced CT image is generated through weighted processing;

[0151] Based on the reinforced CT image, a reconstructed simulation CT image with different halo artifact degrees is generated through adjustment of the coefficient and the cut-off frequency optimization according to the smoothing filter kernel function and the sharp filter kernel function processing.

[0152] The smoothing filter kernel function is:

[0153]

[0154] Wherein, the H'(f) is the frequency response after smoothing filter kernel processing, the f m is the cut-off frequency, and the f is the original frequency;

[0155] The sharp filter kernel function is:

[0156]

[0157] The H''(f) is the frequency response after sharp filter kernel processing, the a is a first weighting coefficient, and the b is a second weighting coefficient.

[0158] Since the reconstruction algorithm of the actual CT is proprietary to each manufacturer, the specific technical parameters are not disclosed, therefore, the four reconstruction kernels (Soft 1 / 2 / 3 and Sharp) from smoothing to sharpness are designed by the application and applied to the reconstruction of the simulation CT projection image. The specific steps are as follows:

[0159] I. Input parameters

[0160] -Original projection data P. The projection data (sinogram) needs to be a three-dimensional matrix projections (angle x detector row x detector column), and the unit is line integral attenuation coefficient (i.e. -log(I / I0)).

[0161] -Geometric parameters:

[0162] - Distance of the source to the detector D(SID).

[0163] - Distance of the source to the phantom R(SAD).

[0164] - Detector pixel size (du, dv).

[0165] - Detector center position (u0, v0).

[0166] II. Weighting (Cosine Weighting)

[0167] - Purpose: To compensate the projection intensity non-uniformity caused by the difference of the ray angle in fan-beam and cone-beam geometry.

[0168]

[0169] o θ is the angle between the ray and the detector plane,

[0170] • Parameters:

[0171] Cone angle parameter v: affects the weight in the vertical direction (z-axis).

[0172] III. Filtering - Purpose: To eliminate the low frequency blur in the projection data and enhance the high frequency edge information; By adjusting the parameters such as the coefficient, the cut-off frequency, etc., the coronary CT images with different degrees of halo artifacts are obtained.

[0173] - Steps:

[0174] - One-dimensional Fourier transform: Perform FFT on the weighted projection data along the detector row direction to obtain the frequency domain data.

[0175] - Frequency domain filtering: Multiply by the ramp filter (Ram-Lak filter) or the windowed filter (such as Shepp-Logan, Hamming window). Ram-Lak preserves all high frequency components (sharp edges, but noise sensitive). Hann / Shepp-Logan suppresses high frequency noise and reduces resolution (similar to low-pass filtering).

[0176] In the present application, for the smoothing filter kernel 1-3, the smoothing filter kernel 1-3 is defined by modifying the cut-off frequency of the Hanning window. In the frequency domain, the Hanning window is defined as follows:

[0177]

[0178] where f m represents the cut-off frequency.

[0179] In the present invention, the frequency response of the sharp filter kernel is defined by the following equation:

[0180]

[0181] where f m represents the cutoff frequency; the preferred value of a is -0.6; the preferred value of b is 1.3.

[0182] As Fig. 6 shown, different window functions of the reconstruction filter: blue (smooth filter kernel 1): function with cutoff frequency at 20% Nyquist; orange (smooth filter kernel 2): function with cutoff frequency at 30% Nyquist; lemon yellow (smooth filter kernel 3): function with cutoff frequency at 100% Nyquist; purple (sharp filter kernel): function with cutoff frequency at 190% Nyquist.

[0183] Inverse Fourier transform: convert the filtered data back to the spatial domain to obtain the filtered projections.

[0184] Parameters:

[0185] Cutoff frequency d: controls the bandwidth of the filter (to avoid amplification of high-frequency noise).

[0186] Window function type: affects the noise and resolution trade-off of the reconstructed image.

[0187] Four, back projection

[0188] Purpose: to map the filtered projection data back to the three-dimensional image space.

[0189] Five, image post-processing

[0190] Optional steps:

[0191] Noise suppression: use Gaussian filtering for denoising. In the present invention, the preferred value of the Gaussian filter is 0.8.

[0192] Truncation correction: handle the truncation artifacts caused by the finite size of the detector (e.g. Parker weight compensation).

[0193] Parameters: filter kernel size, noise model parameters, etc.

[0194] The halo artifact of calcified plaque is related to many factors, and the present application mainly focuses on tube voltage and reconstruction kernel. SPEKTR is used to calculate the X-ray energy spectrum, and the beam voltage is set to 80 / 100 / 120 kV. Considering the balance between image quality and simulation operation time, the number of photons for each simulation is set to 6x108. The simulated projection is generated using GATE (v9.1), which provides accurate CT simulation and comprehensive physical processes. During the simulation imaging process, a water ball is set, and the background is set to air to calibrate the CT value.

[0195] The simulated projection is reconstructed based on the FDK algorithm implemented by TIGRE (image reconstruction toolbox of MATLAB). Similar to the actual CT reconstruction process, a Gaussian filter is applied in the projection domain to suppress noise (SD=0.8). Since the reconstruction algorithm of actual CT is proprietary to each manufacturer, the technical parameters are not disclosed, so the present application designs four reconstruction kernels from soft to sharp (Soft1 / 2 / 3 and Sharp) for application in simulated images.

[0196] The visualization optimization includes:

[0197] The visualization optimization function is:

[0198]

[0199] The μ is the linear attenuation coefficient, the u_water is the attenuation value of water, and the u_object is the attenuation value of the object, wherein the object is a digital heart coronary plaque phantom, and the CT Attenuation Value is the simulated CT data after visualization optimization processing.

[0200] In the present application, the gray scale value of the reconstructed image reflects the linear attenuation coefficient (μ) of the material, which can be converted to the CT value (unit HU) by the formula. In addition, the visualization result can be obtained by the drawing command.

[0201]

[0202] By adjusting the definition of the material (especially the calcified plaque), different material CT values can be obtained; by adjusting the shape of the plaque, different shapes of calcified / lipid / fibrous plaque shapes can be obtained; by adjusting the reconstruction filter kernel, tube voltage, and size of the detector, different sizes of calcified plaque halo artifacts can be obtained.

[0203] A multi-parameter model for generating halo artifacts of simulated calcified plaques is proposed, which includes the combination of the sharpness of the reconstruction filter kernel, the size of the tube voltage, and the size of the detector, which can generate halo artifacts of different sizes of calcified plaques, solving the problem of fixed plaque shape in traditional CT simulation; wherein, based on the smoothness and sharpness of the reconstruction filter kernel, the filter kernel is designed, and the degree of halo artifact is adjusted by parameters to further simulate the halo artifact.

[0204] The simulation method further comprises:

[0205] Based on the simulated CT images of different virtual lesions, the simulated CT samples are generated through dataset samples, and finally used as the training samples S106 S1006 for the current halo artifact removal training.

[0206] Through the generation of the above-mentioned simulated CT samples, the algorithm generalization is improved (1) diversified dataset: a large number of variant data (such as different plaque shapes + different scan parameter combinations) are generated through the digital phantom, which significantly improves the adaptability of the deep learning algorithm to unknown cases. (2) controllable extreme cases: rare but clinically important extreme cases (such as severe stenosis with microcalcification) can be generated automatically, which makes up for the problem of insufficient real data.

[0207] The heart coronary plaque simulation system based on the digital phantom comprises:

[0208] Based on the simulation method according to any one of the above, the simulation system comprises a simulation platform, and the simulation platform comprises:

[0209] A heart coronary artery base model generation module: based on clinical heart coronary artery data, different lesion degrees of heart coronary artery base models are generated through automatic region labeling or manual modeling S101 S1001, wherein the different lesion degrees include different stenosis degrees, different plaque shapes, and different plaque material lesion degrees;

[0210] A digital heart coronary plaque phantom module: based on the heart coronary artery base model, digital heart coronary plaque phantoms of different material tissues are generated through material parameter setting S102 S1002, wherein the different material tissues include coronary artery, contrast agent, calcified plaque, fibrous tissue, and lipid tissue;

[0211] A simulated CT projection image generation module S103: based on the digital heart coronary plaque phantom, simulated CT projection images are generated through simulated CT imaging according to the simulation environment setting S103 S1003; wherein, the simulation environment setting includes radiation source simulation setting, tube voltage and its commonly used range simulation setting, photon emission number simulation setting of each projection angle, detector pixel size setting, and detector imaging matrix setting;

[0212] A reconstructed simulation CT projection image module: based on the simulation CT projection image, a reconstructed simulation CT projection image S104 S1004 with different halo artifacts is generated by a reconstruction filter kernel, wherein the reconstruction filter kernel includes a construction smoothing filter kernel and a sharp filter kernel;

[0213] A judgment of the current halo artifact removal algorithm accuracy module: based on the reconstructed simulation CT projection image, the algorithm de-artifact result S105 is obtained by introducing it into the current halo artifact removal algorithm, and the algorithm de-artifact result and different virtual lesion setting errors are judged, and then the current halo artifact removal algorithm accuracy S1005 is judged.

[0214] 1. Hardware optimization: the invention can cover the CT device parameter recommendation system: according to the simulation results, the tube voltage / reconstruction kernel combination is automatically recommended to reduce the calcification plaque halo artifact in the real machine CT (for example, the prompt “120kVp+smooth kernel” is suitable for small calcification focus).

[0215] 2. Clinical software tool: (1) Halo artifact prediction plug-in: provide SDK integration service to CT manufacturers, input scan parameters to preview artifact morphology, and assist technicians to optimize scan scheme. (2) Patient-specific vessel reconstruction technology: quickly generate personalized digital vessel model from clinical images, and retain the topological characteristics of calcification / lipid distribution. (3) Plaque evolution simulation: dynamic evolution model of lipid core.

[0216] 3. AI algorithm training application: (1) Synthetic data generation: a method for generating labeled CT training set, which generates calcification images with different artifact intensities in batches through calculation model, and labels the real calcification area. Solve the problem of insufficient real data and high labeling cost in AI training. (2) De-artifact algorithm framework: a physics-based generative adversarial network (Physics-GAN), which uses the mathematical model of simulation artifact as the constraint condition of the generator. Compared with traditional pure data-driven de-artifact network, the generalization is improved. (3) Evaluation benchmark construction: propose a standardized test set: containing simulation artifact image, entity CT image and real calcification mask triple data, as the gold standard test platform of de-artifact algorithm.

[0217] Experiment one:

[0218] As Fig. 3As shown, in this experiment, a total of 20 computational models (STL format) were designed to simulate vascular segments with different degrees of stenosis, including the coronary artery wall, lumen, fibrous tissue, lipid plaque and calcified plaque. The geometric shape of each model along the axial direction is consistent. The length, wall thickness and cross-sectional outer diameter of the vascular model are 15, 1 and 8 mm, respectively. In order to represent the variability of plaques, 5 models were defined, named Model 1-5. The total plaque area (including fibrous tissue, lipid plaque and calcified plaque) was set to 10.33 mm2, 15.75 mm2, 19.40 mm2 and 23.82 mm2, respectively, corresponding to four degrees of stenosis (mild, moderate, advanced and severe), and the area stenosis percentages were 36.54%, 55.71%, 68.62% and 84.26%. The area of ​​calcified plaques ranged from 4.95 to 23.82 mm2;

[0219] exist Fig. 3 Middle, A: Volume rendering. B: Central cross-section. C: Models of different shapes and stenosis degrees. In Figure C, from left to right: Models 1-5; from top to bottom: mild, moderate, advanced, and severe stenosis. Note: Pink represents the vessel wall, yellow represents lipid plaque, green represents fibrous tissue, white represents calcified plaque, and gray represents the contrast agent-filled coronary artery lumen.

[0220] The composition of calcified plaque was defined as dense calcified plaque with a CT value range of 500–1200 HU (at 100 kV). Material definitions for fibrous tissue and lipid plaque were modified from the GateMaterials.db (GATE materials database) material definitions for fat and muscle to match the CT values ​​of fibrous tissue (131–350 HU) and lipid plaque (−42–47 HU). The composition of the contrast agent was defined based on the molecular formula of iodine iodide (XENETIX 350).

[0221] The halo artifact of calcified plaque is related to many factors. The present invention mainly focuses on the tube voltage and reconstruction kernel. The X-ray energy spectrum was calculated using SPEKTR, and the beam voltage was set to 80 / 100 / 120 kV.

[0222] Utilizing the technical solution of the present invention, or those skilled in the art designing similar technical solutions inspired by the technical solution of the present invention to achieve the above-mentioned technical effects, all fall within the scope of protection of the present invention. It should be noted that the specific implementation methods in the embodiments of the present invention can be at least partially combined with each other, and the combination and use of the above-mentioned embodiments, or their different combinations, should also be considered to fall within the scope of protection of the present invention.

Claims

1. A method for simulating coronary artery plaque based on a digital phantom, characterized in that: include: Based on clinical coronary artery data, a basic model of the coronary artery with different degrees of lesions is generated by automatic region annotation or manual modeling, wherein the different degrees of lesions include lesions with different degrees of stenosis, different plaque shapes, and different plaque materials; Based on the basic model of the coronary arteries, digital coronary artery plaque phantoms with different material tissues are generated by setting material parameters. The different material tissues include coronary artery vessels, contrast agents, calcified plaques, fibrous tissue, and lipid tissue. Based on a digital heart coronary artery plaque phantom, simulated CT projection images are generated through simulated CT imaging according to simulation environment settings; wherein the simulation environment settings include simulation settings of the radiation source, simulation settings of the tube voltage and its common range, simulation settings of the number of photons emitted at each projection angle, settings of the detector pixel size, and settings of the detector imaging matrix; Based on the simulated CT projection image, reconstructed simulated CT projection images with different halo artifact levels are generated by reconstructing a filter kernel, wherein the reconstruction filter kernel includes a smoothing filter kernel and a sharpening filter kernel; Based on the reconstructed simulated CT projection image, the algorithm artifact removal result is obtained by importing it into the current halo artifact removal algorithm, and the error between the algorithm artifact removal result and different virtual lesion settings is judged, and then the accuracy of the current halo artifact removal algorithm is judged.

2. The simulation method according to claim 1, wherein: The automatic area marking includes: Based on clinical coronary artery data, noise reduction is performed through median filtering to generate denoised vascular data; Generate vascular edge features by fitting the vascular edge based on the denoised vascular data; Based on the edge features of blood vessels, high-contrast blood vessel images are generated through histogram equalization enhancement; Based on the high-contrast vascular image, a high-contrast vascular grayscale image is generated by image weighted averaging; Based on the high-contrast vascular grayscale image, an annotated vascular image is generated according to the target annotation rule based on the image CT value; Based on the annotated vascular image, generating a binary vascular image having a determined annotated region through binarization processing, wherein the determined annotated region includes the heart, the coronary artery, and the calcified plaque; Based on the binary vascular image of the determined marked area, a three-dimensional model file corresponding to the digital heart coronary artery plaque phantom is generated by extracting the marked area.

3. The simulation method according to claim 2, characterized in that: The target labeling rules based on image CT values ​​include: If the static image CT value of the current region is 30-50HU, the current region is marked as myocardial tissue; If the static image CT value of the current region is 40-60HU, the current region is initially judged as vascular tissue. If the current region is initially judged as vascular tissue, if the blood flow image CT value of the current region is 30-50HU, the current region is marked as a coronary artery. If the static image CT value of the current area is ≥130HU and the image edge is sharp and appears as high-density spots or flakes, the current area is initially judged as a typical calcified plaque. If the current area is initially judged as a typical calcified plaque, further judgment is made: If the static image CT value of the current area is 130-199HU, the current area is marked as a mild calcified plaque; If the static image CT value of the current area is 200-299HU, the current area is marked as moderate calcified plaque; If the static image CT value of the current area is ≥300HU, the current area is marked as a severely calcified plaque.

4. The simulation method according to claim 1, wherein: The simulated CT imaging includes: Based on the digital heart coronary artery plaque phantom data, pre-processed phantom data is generated through phantom data generation; Generate simulation environment settings through simulation parameter settings based on basic parameters of radiation source, basic parameters of tube voltage, and basic parameters of photon emission; Based on the pre-processed phantom data and simulation environment settings, simulated CT projection images are generated through virtual rotation scanning.

5. The simulation method according to claim 4, characterized in that: The phantom data generation includes: Based on the digital heart coronary artery plaque phantom data, a digital phantom element dataset is generated through bioinformatics analysis; Based on the digital phantom dataset, a low-noise phantom dataset is generated by filtering low-quality data; Based on the low-noise phantom dataset, a standard phantom dataset with different annotated regions is generated by optimizing display contrast, wherein the different annotated regions include the heart, coronary arteries, and calcified plaques; The simulation parameter settings include: Based on the basic parameters of the radioactive source, the radioactive source simulation settings are generated through the radioactive source simulation parameter settings; Based on the basic parameters of tube voltage, the simulation settings of tube voltage and its common range are generated through the tube voltage simulation parameter settings; Based on the basic parameters of photon emission, the simulation settings of the number of photon emissions at the current projection angle are generated through the photon emission simulation parameter settings; Based on the simulation settings of the radiation source, the tube voltage and its common range, and the number of photon emissions per projection angle, they are aggregated into simulation parameter settings; The virtual rotation scan includes: Based on a preset angle sequence, the detector is rotated to a preset angle by virtually sending a pulse signal, wherein the preset angle sequence includes, for example, 0°, 180°, 360°, and 0°; Based on the pre-processed phantom data and according to the simulation environment settings, the detector is rotated and scanned in a spiral CT virtual mode, wherein the spiral CT virtual mode includes continuous rotation with a step size of ≤0.1 and a 360° rotation coverage; When the detector rotates to the preset angle and stabilizes, a virtual X-ray exposure is performed and the detector data acquisition thread is started to generate a virtual exposure image at the current angle; Based on the virtual exposure image of the current angle, a multi-angle simulated CT projection image set containing metadata such as angle tags and timestamps is generated by combining exposure images.

6. The simulation method according to claim 1, characterized in that: The reconstruction filter kernel includes: Based on the simulated CT projection image, an enhanced CT image is generated through weighted processing; Based on the enhanced CT image, the smoothing filter kernel function and the sharpening filter kernel function are processed, and the coefficients and the cutoff frequency are adjusted to optimize the reconstructed simulated CT projection images with different halo artifact degrees.

7. The simulation method according to claim 6, characterized in that: include: The smoothing filter kernel function: Wherein, H'(f) is the frequency response after smoothing filter kernel processing, and f m is the cutoff frequency, and f is the original frequency; The sharp filter kernel function: The H"(f) is the frequency response after sharp filter kernel processing, a is the first weighting coefficient, and b is the second weighting coefficient.

8. The simulation method according to claim 1, wherein: The reconstruction of the filter kernel also includes visualization optimization, which includes: Visual optimization function: The μ is a linear attenuation coefficient, the u_water is the attenuation value of water, and the u_object is the attenuation value of an object, wherein the object is a digital heart coronary artery plaque phantom, and the CT Attenuation Value is simulated CT data after visualization optimization processing.

9. The simulation method according to claim 1, characterized in that: The simulation method further comprises: Based on the simulated CT projection images of different virtual lesions, simulated CT samples are generated through data set sample construction, and finally used as training samples for the current halo artifact removal training.

10. A coronary artery plaque simulation system based on a digital phantom, characterized in that: include: Based on the simulation method according to any one of claims 1 to 9, the simulation system includes a simulation platform, and the simulation platform includes: Generate a coronary artery basic model module: Based on clinical coronary artery data, generate coronary artery basic models with different lesion degrees through automatic region annotation or manual modeling, wherein the different lesion degrees include lesion degrees with different stenosis degrees, different plaque shapes, and different plaque materials; Generate digital coronary plaque phantom module: Based on the basic coronary artery model, generate digital coronary plaque phantoms of different material tissues by setting material parameters. The different material tissues include coronary artery vessels, contrast agents, calcified plaques, fibrous tissue, and lipid tissue. Generate simulated CT projection image module: Based on the digital heart coronary artery plaque phantom, according to the simulation environment settings, generate simulated CT projection images through simulated CT imaging; wherein the simulation environment settings include the simulation settings of the radiation source, the simulation settings of the tube voltage and its common range, the simulation settings of the number of photons emitted at each projection angle, the settings of the detector pixel size, and the settings of the detector imaging matrix; A module for generating reconstructed simulated CT projection images: generating reconstructed simulated CT projection images with different halo artifact levels based on the simulated CT projection images by reconstructing filter kernels, wherein the reconstruction filter kernels include a smoothing filter kernel and a sharpening filter kernel; Module for judging the accuracy of the current de-halo artifact algorithm: Based on the reconstructed simulated CT projection image, by importing it into the current de-halo artifact algorithm, the algorithm de-halo artifact result is obtained, and the error between the algorithm de-halo artifact result and different virtual lesion settings is judged, and then the accuracy of the current de-halo artifact algorithm is judged.

Images