Photon-counting ct high-resolution scan mode for quantitative assessment of ct-ffr

Abstract

The application discloses a CT-FFR quantitative evaluation system based on a high-resolution photon CT scanning mode, belongs to the technical field of coronary artery stenosis, and comprises the following modules: a data acquisition and image generation module, which is used for acquiring coronary artery photon CT image data; a three-dimensional reconstruction and fluid analysis module, which is used for processing and three-dimensionally reconstructing the coronary artery photon CT image data, generating a three-dimensional geometric model of a coronary artery tree and a stenosis section, and outputting an FFR value of an arbitrary position of the coronary artery; and a CT-FFR precise quantitative evaluation module, which is used for analyzing high-resolution coronary artery photon CT image data and determining a CT-FFR precise quantitative evaluation result. The application solves the problems that the existing coronary artery function evaluation cannot be high-precision, non-invasive and efficient, and the coronary artery flow reserve fraction cannot be precisely quantified. The application can realize high-precision, non-invasive and efficient coronary artery function evaluation and can realize precise quantification of the coronary artery flow reserve fraction.

Description

Technical Field

[0001] This invention relates to the field of coronary artery stenosis technology, specifically a quantitative assessment system for CT-FFR using high-resolution photon CT scanning mode. Background Technology

[0002] Coronary artery stenosis is a narrowing of the blood vessel lumen caused by atherosclerosis. In severe cases, it can lead to coronary heart disease. Diagnosis requires coronary angiography to determine the degree of stenosis. Treatment methods include drug therapy, interventional stent implantation, and coronary artery bypass surgery. Among these, the fractional flow reserve (FFR) is an important indicator for assessing the physiological function of coronary artery stenosis. It reflects the ratio of the blood flow to the myocardial region supplied by the distal coronary artery under maximal congestion to the maximum blood flow that the region can normally obtain.

[0003] Existing technologies cannot achieve high-precision, non-invasive, and efficient coronary functional assessment, nor can they accurately quantify the fractional flow reserve of the coronary arteries. Therefore, early screening for coronary artery stenosis lacks functional basis. Summary of the Invention

[0004] The purpose of this invention is to provide a quantitative assessment system for CT-FFR in high-resolution photonic CT scanning mode, which can achieve high-precision, non-invasive and efficient coronary artery functional assessment, accurately quantify coronary blood flow reserve, and provide reliable functional basis for early screening of coronary artery stenosis, thus solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A quantitative evaluation system for CT-FFR in high-resolution photon CT scanning modes includes:

[0007] The data acquisition and image generation module is used to acquire high-resolution coronary artery photon CT image data based on photon CT.

[0008] The 3D reconstruction fluid analysis module is used to process and 3D reconstruct high-resolution coronary artery photon CT image data, generate 3D geometric models of the coronary artery tree and stenotic segments, and output the FFR value at any location of the coronary artery.

[0009] The CT-FFR precise quantitative assessment module is used to construct a CT-FFR quantitative assessment model based on artificial intelligence and computational fluid dynamics analysis, and to analyze and recognize patterns in high-resolution coronary photon CT image data to determine the CT-FFR precise quantitative assessment results.

[0010] Preferably, high-resolution coronary artery photon CT image data acquired based on photon CT includes:

[0011] Photon CT is used to monitor the coronary arteries of patients. It uses a photon counting detector to directly convert X-ray photons into electrical signals, captures and analyzes the energy information of X-ray photons, distinguishes different material components, clearly displays the edge of the coronary artery lumen and the calcified plaque components, and obtains high-resolution coronary artery photon CT image data.

[0012] Preferably, the three-dimensional reconstruction fluid analysis module includes:

[0013] The image processing unit is used to process the coronary artery photon CT image data.

[0014] Among them, noise reduction processing of coronary artery photonic CT image data is performed based on filters to remove noise from the coronary artery photonic CT image data, improve the signal-to-noise ratio and eliminate inherent hardware errors;

[0015] Examine the coronary artery photon CT image data to identify missing and outlier values.

[0016] To evaluate missing and outlier values ​​in coronary photonic CT image data and determine whether missing and outlier values ​​in coronary photonic CT image data are valuable for quantitative assessment of CT-FFR;

[0017] When missing and outlier values ​​in coronary photon CT image data are valuable for quantitative assessment of CT-FFR, missing values ​​are filled and outlier values ​​are replaced.

[0018] Radiographic hardening correction is used to correct coronary artery photon CT image data, reducing and eliminating radiographic hardening artifacts in the data.

[0019] Preferably, the three-dimensional reconstruction fluid analysis module further includes:

[0020] The 3D modeling unit is used to construct the 3D geometric model of the coronary artery tree and stenotic segments and to perform computational fluid dynamics analysis.

[0021] Specifically, coronary artery photon CT image data is reconstructed into three-dimensional coronary tree volume data, and image segmentation algorithms are used to segment the three-dimensional coronary tree volume data. The morphology of the coronary tree is accurately delineated from the three-dimensional coronary tree volume data, stenotic segments are identified, and the segmentation results are determined.

[0022] Based on computational fluid dynamics and combined with the segmentation results, a three-dimensional geometric model of the coronary artery tree and the stenotic segment is constructed, while preserving the geometric features of the stenotic segment;

[0023] Boundary conditions are set and dynamic blood flow simulation is performed. The inlet boundary is the pressure or flow rate at the aortic root, and the outlet boundary is set based on the microcirculation resistance model of the myocardium downstream of the coronary artery. The vessel wall is a rigid wall. The blood flow pressure field is calculated and the FFR value at any location of the coronary artery is output.

[0024] Preferably, the three-dimensional coronary tree volume data is segmented using an image segmentation algorithm, including:

[0025] The reconstructed 3D coronary tree volume data is retrieved, and the 3D coronary tree volume data is preprocessed, including noise removal and enhancement of vascular features.

[0026] Data extraction is performed on the preprocessed coronary tree volume data to obtain the initial central axis of the coronary tree.

[0027] Determine the shortest distance between each voxel contained in the coronary tree volume data and the initial central axis of the coronary tree;

[0028] The average radius of the main trunk and branches of the coronary artery tree and the average angle between the branches and the main trunk were statistically analyzed, and the structural constraint factors of fusion distance attenuation and branch angle penalty for each voxel were constructed.

[0029] Three types of features are extracted from the three-dimensional coronary tree volume data after data preprocessing. The weights of the three types of features are dynamically allocated through structural constraint factors. The fused feature values ​​are obtained by weighted summation of the three types of features and their corresponding weights.

[0030] An adaptive threshold is set based on the fused feature values ​​to distinguish between vascular and non-vascular regions, generating a segmentation mask M. init (x,y,z); where the three types of features include grayscale feature F1, gradient feature F2 that highlights the edge of the blood vessel, and simplified vascularness feature F3 that reflects the uniformity of the blood vessel structure;

[0031] Using the segmentation mask M init (x,y,z) extracts three-dimensional volume data containing only the vascular region from the original three-dimensional coronary tree volume data to complete the coronary tree volume data segmentation.

[0032] Preferably, three types of features are extracted from the preprocessed 3D coronary tree volume data, and the weights of the three types of features are dynamically assigned through structural constraint factors, including:

[0033] The structural constraint factor is retrieved and compared with a preset first factor reference value and a second factor reference value; wherein the first factor reference value is 0.7 and the second factor reference value is 0.4.

[0034] When the structural constraint factor exceeds the reference value of the first factor, it is determined that the voxel corresponding to the structural constraint factor is located in the core vascular region.

[0035] When the structural constraint factor does not exceed the first factor reference value but exceeds the second factor reference value, it is determined that the voxel corresponding to the structural constraint factor is in the edge region or branch region.

[0036] When the structural constraint factor does not exceed the reference value of the second factor, the voxel corresponding to the structural constraint factor is determined to be in the transition region.

[0037] Preferably, the CT-FFR precise quantitative assessment module includes:

[0038] The model building unit is used to construct a quantitative assessment model for CT-FFR based on artificial intelligence and computational fluid dynamics analysis.

[0039] Among them, the correspondence between the three-dimensional geometric model of the coronary tree and the stenotic segment and the FFR value at any position of the coronary artery calculated by computational fluid dynamics is monitored, and historical data of the coronary artery blood flow pressure field are collected.

[0040] The historical data of coronary artery blood flow pressure field were divided into training set and test set in a 7:3 ratio.

[0041] The deep learning model is trained using a training set, enabling it to autonomously learn the quantitative assessment behavior of CT-FFR from the training set and accurately quantify the fractional coronary flow reserve, thus determining the quantitative assessment model of CT-FFR.

[0042] The CT-FFR quantitative assessment model was tested using a test set, and its generalization performance was evaluated. The model test evaluation results were determined, and the CT-FFR quantitative assessment model was optimized based on the model test evaluation results to determine the optimal CT-FFR quantitative assessment model.

[0043] Preferably, the CT-FFR quantitative assessment model is tested using a test set, and the generalization performance of the CT-FFR quantitative assessment model is evaluated by performing the following operations:

[0044] The test set is input into the CT-FFR quantitative assessment model. The CT-FFR quantitative assessment model is tested based on the test set to determine whether the CT-FFR quantitative assessment model can achieve the expected effect of accurately quantifying the coronary blood flow reserve.

[0045] When the CT-FFR quantitative assessment model fails to achieve the expected effect of accurately quantifying the fractional flow reserve of the coronary arteries, the parameters of the CT-FFR quantitative assessment model are adjusted and iteratively optimized until the CT-FFR quantitative assessment model can achieve the expected effect of accurately quantifying the fractional flow reserve of the coronary arteries, and the optimal CT-FFR quantitative assessment model is determined.

[0046] Preferably, the CT-FFR precise quantitative assessment module further includes:

[0047] The precision quantification unit is used to precisely quantify the fractional flow reserve of the coronary arteries.

[0048] The process involves inputting high-resolution coronary artery photon CT image data into a CT-FFR quantitative assessment model. The model analyzes and performs pattern recognition on the high-resolution coronary artery photon CT image data, and accurately quantifies the fractional flow reserve of the coronary arteries to determine the precise quantitative assessment result of CT-FFR.

[0049] Preferably, the CT-FFR precise quantitative assessment module further includes:

[0050] The results output unit is used to output and display the precise quantitative assessment results of CT-FFR in a visual format.

[0051] Among them, the coronary artery stenosis myocardial ischemia was assessed based on the precise quantitative assessment results of CT-FFR, with FFR≤0.80 as the critical value for judging significant functional ischemia, so as to achieve quantitative assessment of the severity of coronary artery blood flow obstruction in patients.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] This invention utilizes photonic CT to clearly display the edges of the coronary artery lumen and the composition of calcified plaques. It acquires high-resolution coronary photonic CT image data, processes and reconstructs this data to generate a three-dimensional geometric model of the coronary artery tree and stenotic segments, sets boundary conditions, performs dynamic blood flow simulation, calculates the blood flow pressure field, and outputs the FFR value at any location in the coronary artery. Based on artificial intelligence and computational fluid dynamics analysis, a CT-FFR quantitative assessment model is constructed. This model is then used to analyze and perform pattern recognition on the high-resolution coronary photonic CT image data, accurately quantifying the coronary fractional flow reserve. This precise quantitative assessment of CT-FFR results enables high-precision, non-invasive, and efficient coronary functional assessment, achieving accurate quantification of the coronary fractional flow reserve. Attached Figure Description

[0054] Figure 1This is a block diagram of the quantitative evaluation system for CT-FFR using the high-resolution scanning mode of photonic CT according to the present invention.

[0055] Figure 2 This is a flowchart of the quantitative evaluation system for CT-FFR using the high-resolution scanning mode of photonic CT according to the present invention. Detailed Implementation

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

[0057] To address the limitations of current methods that fail to provide high-precision, non-invasive, and efficient coronary functional assessment, accurate quantification of coronary flow reserve, and reliable functional evidence for early screening of coronary artery stenosis, please refer to [link to relevant documentation]. Figures 1-2 This embodiment provides the following technical solution:

[0058] The quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode includes: a data acquisition and image generation module, a three-dimensional reconstruction and fluid analysis module, and a precise quantitative evaluation module for CT-FFR.

[0059] Specifically, through the interactive communication between the data acquisition and image generation module, the three-dimensional reconstruction fluid analysis module, and the CT-FFR precise quantitative assessment module, high-precision, non-invasive, and efficient coronary artery functional assessment can be achieved. It can accurately quantify the fractional flow reserve of the coronary arteries and provide reliable functional evidence for the early screening of coronary artery stenosis.

[0060] The data acquisition and image generation module is used to acquire high-resolution coronary artery photon CT image data based on photon CT.

[0061] In this embodiment, high-resolution coronary artery photon CT image data is acquired based on photon CT, including:

[0062] Photon CT is used to monitor the coronary arteries of patients. It uses a photon counting detector to directly convert X-ray photons into electrical signals, captures and analyzes the energy information of X-ray photons, distinguishes different material components, clearly displays the edge of the coronary artery lumen and the calcified plaque components, and obtains high-resolution coronary artery photon CT image data.

[0063] It should be noted that photon CT is an advanced medical imaging technology based on photon counting detectors. By directly capturing and analyzing the energy information of X-ray photons, it significantly improves image resolution and diagnostic accuracy. Compared with traditional CT, it can distinguish X-ray photons of different energy levels, achieving clearer tissue contrast and lower radiation dose, and has great potential in fields such as early tumor screening and cardiovascular imaging.

[0064] Traditional CT uses indirect conversion detectors, such as scintillators and photodiodes, to convert X-ray photons into visible light before generating electrical signals. This process involves energy information loss and noise interference. In contrast, photon-counting CT uses semiconductor materials to directly record the energy value of a single X-ray photon. By using energy thresholds, it distinguishes different tissues or substances, thus providing multi-parameter imaging capabilities. Photon CT offers the following advantages:

[0065] 1) High resolution: Photon CT has a spatial resolution of up to 0.2 mm, which can clearly display tiny structures, such as the fine anatomical features of the coronary arteries; 2) Low radiation dose: Compared with traditional CT, photon CT has a significantly lower radiation dose, making it more suitable for patients who need to undergo frequent imaging examinations; 3) Multi-spectral imaging: Through energy resolution technology, photon CT can distinguish different material components and provide richer anatomical and functional information.

[0066] Therefore, high-resolution coronary artery photon CT image data acquired by photon CT can effectively reduce artifacts and more clearly display the edges of the coronary artery lumen and plaque components such as calcification, providing a data foundation for subsequent high-precision analysis.

[0067] The three-dimensional reconstruction fluid analysis module is used to process and reconstruct high-resolution coronary artery photon CT image data, generate a three-dimensional geometric model of the coronary artery tree and stenotic segment, and output the FFR value at any location of the coronary artery.

[0068] In this embodiment, the three-dimensional reconstruction fluid analysis module includes:

[0069] The image processing unit is used to process the coronary artery photon CT image data.

[0070] Among them, noise reduction processing of coronary artery photonic CT image data is performed based on filters to remove noise from the coronary artery photonic CT image data, improve the signal-to-noise ratio and eliminate inherent hardware errors;

[0071] Examine the coronary artery photon CT image data to identify missing and outlier values.

[0072] To evaluate missing and outlier values ​​in coronary photonic CT image data and determine whether missing and outlier values ​​in coronary photonic CT image data are valuable for quantitative assessment of CT-FFR;

[0073] When missing and outlier values ​​in coronary photon CT image data are valuable for quantitative assessment of CT-FFR, missing values ​​are filled and outlier values ​​are replaced.

[0074] X-ray hardening correction is used to correct coronary artery photon CT image data, reducing and eliminating X-ray hardening artifacts in the data. X-ray beams contain photons of different energies, and low-energy photons are more easily absorbed, resulting in higher average beam energy and hardening. This can produce artifacts at the edges of high-density objects such as bones and calcifications. By correcting the coronary artery photon CT image data, these artifacts can be corrected more accurately.

[0075] It should be noted that image quality can be improved by performing noise reduction and correction operations on coronary artery photon CT image data.

[0076] In this embodiment, the three-dimensional reconstruction fluid analysis module further includes:

[0077] The 3D modeling unit is used to construct the 3D geometric model of the coronary artery tree and stenotic segments and to perform computational fluid dynamics analysis.

[0078] Specifically, coronary artery photon CT image data is reconstructed into three-dimensional coronary tree volume data, and image segmentation algorithms are used to segment the three-dimensional coronary tree volume data. The morphology of the coronary tree is accurately delineated from the three-dimensional coronary tree volume data, stenotic segments are identified, and the segmentation results are determined.

[0079] Based on computational fluid dynamics and combined with the segmentation results, a three-dimensional geometric model of the coronary artery tree and the stenotic segment is constructed, while preserving the geometric features of the stenotic segment;

[0080] Boundary conditions are set and dynamic blood flow simulation is performed. The inlet boundary is the pressure or flow rate at the aortic root, and the outlet boundary is set based on the microcirculation resistance model of the myocardium downstream of the coronary artery. The vessel wall is a rigid wall. The blood flow pressure field is calculated and the FFR value at any location of the coronary artery is output.

[0081] The CT-FFR precise quantitative assessment module is used to construct a CT-FFR quantitative assessment model based on artificial intelligence and computational fluid dynamics analysis, and to analyze and recognize patterns in high-resolution coronary photon CT image data to determine the CT-FFR precise quantitative assessment results.

[0082] It should be noted that the process involves processing and 3D reconstruction of coronary artery photon CT image data. Based on the segmentation results, a 3D geometric model of the coronary artery tree and stenotic segment is generated. Boundary conditions are set, the blood flow pressure field is calculated, and the FFR value at any location in the coronary artery is output. A deep learning model is used to quickly predict FFR, replacing some computational fluid dynamics calculations and improving efficiency. Based on high-resolution photon CT coronary CT angiography, combined with artificial intelligence and computational fluid dynamics, accurate quantification of coronary flow reserve (CT-FFR) is achieved, enabling non-invasive assessment of whether coronary artery stenosis leads to myocardial ischemia. Through the advantages of PCCT hardware, the integration of artificial intelligence and computational fluid dynamics algorithms, and full-process automation, high-precision, non-invasive, and efficient coronary functional assessment is achieved.

[0083] In this embodiment, an image segmentation algorithm is used to segment the three-dimensional coronary tree volume data, including:

[0084] The reconstructed 3D coronary tree volume data was retrieved and preprocessed. This preprocessing included noise reduction and vascular feature enhancement. Specifically, noise reduction employed fast guided filtering to remove CT image noise while preserving vessel edges, with a filter window set to 3×3×3 (balancing efficiency and effectiveness). Vascular feature enhancement primarily involved using a contrast enhancement model to locally enhance the contrast of each voxel in the coronary tree volume data for small branches and stenotic segments, highlighting the differences between the vessels and surrounding tissues.

[0085] Data extraction is performed on the preprocessed coronary tree volume data to obtain the initial central axis of the coronary tree.

[0086] Determine the shortest distance between each voxel contained in the coronary tree volume data and the initial central axis of the coronary tree;

[0087] The average radius of the main trunk and branches of the coronary artery tree and the average angle between the branches and the main trunk were statistically analyzed, and the structural constraint factors of fusion distance attenuation and branch angle penalty for each voxel were constructed.

[0088] The structural constraint factor is obtained by the following formula:

[0089]

[0090] Where V(x, y, z) represents the structural constraint factor, x, y, and z represent the axis letters corresponding to the voxel coordinate axes in the coronary tree volume data, respectively; d(x, y, z) represents the shortest distance from each voxel to the axis; R(x, y, z) represents the average radius; and θ(x, y, z) represents the angle between the voxel's location and the main coronary artery. This represents the average angle between the pre-defined branches and trunk of the coronary artery tree. This structural constraint factor, through the fusion of distance attenuation and angle penalty, forces the segmentation result to conform to the natural anatomical shape of the coronary artery tree, avoiding "inflated / decreased vessel diameter" and "misjudgment of stenosis length" caused by missegmentation of non-vascular tissue or branch breakage. For example, stenosis is easily missed by conventional algorithms due to its low grayscale, but this formula, through the dual constraints of angle and distance, ensures that the boundary of the stenosis is completely delineated, and controls the measurement error of the minimum lumen diameter of the stenosis to within 0.2mm, providing accurate input for "resistance calculation at the stenosis" in the hemodynamic model.

[0091] Three types of features are extracted from the three-dimensional coronary tree volume data after data preprocessing. The weights of the three types of features are dynamically allocated through structural constraint factors. The fused feature values ​​are obtained by weighted summation of the three types of features and their corresponding weights.

[0092] An adaptive threshold is set based on the fused feature values ​​to distinguish between vascular and non-vascular regions, generating a segmentation mask M. init (x,y,z); where the three types of features include grayscale feature F1, gradient feature F2 that highlights the edge of the blood vessel, and simplified vascularness feature F3 that reflects the uniformity of the blood vessel structure;

[0093] Using the segmentation mask M init (x,y,z) extracts three-dimensional volume data containing only the vascular region from the original three-dimensional coronary tree volume data to complete the coronary tree volume data segmentation.

[0094] The aforementioned technical solution integrates the "shortest distance from the voxel to the coronary artery axis" and the "angle between the voxel and the main coronary artery" using a structural constraint factor V(x,y,z). Leveraging the prior knowledge of the coronary tree's "tree-like branches and continuous axes," it forces the segmentation results to conform to the natural morphology of the coronary arteries. This design effectively solves problems such as "mis-segmentation of non-vascular regions" and "coronary artery branch breakage" that easily occur in general segmentation algorithms (such as thresholding methods and region growing methods without structural constraints). It ensures that the three-dimensional morphology of the coronary tree (including small branches and stenotic segments) is completely and accurately delineated, laying a structurally complete foundation for subsequent lesion analysis (such as stenosis identification).

[0095] On the other hand, the above-mentioned technical solution transforms the prior knowledge (distance, angle) of the coronary tree anatomy into quantified structural constraint factors and forms a closed-loop linkage with the dynamic weight allocation of three types of lightweight features. This not only solves the traditional pain points in existing technologies such as "general segmentation algorithms cannot adapt to the tree-like morphology of the coronary tree", "fixed weights are difficult to take into account the differences between the main trunk and small branches / stenotic segments", and "noise and myocardial background interference lead to insufficient segmentation accuracy", but also produces the following technical effects:

[0096] First, the fusion of distance decay and angle penalty by the structural constraint factor not only effectively avoids missegmentation of non-vascular regions, but also unexpectedly improves the continuity and consistency of the stenotic segment boundary. Conventional algorithms are prone to blurring or breaking the boundary due to low grayscale and incomplete structure when processing stenotic segments. However, this scheme, through the dynamic adaptation of constraint factors and gradient features, controls the transition boundary error between stenotic segments and normal blood vessels to within 0.2mm, which is far better than the average error of 0.5mm in the existing technology. Moreover, the quantification accuracy of the length and degree of stenosis of the stenotic segment is improved by more than 40%.

[0097] Second, the synergistic effect of local contrast enhancement and dynamic weight allocation breaks through the cognitive limitation that "lightweight algorithms are difficult to accurately identify small branches". Conventional lightweight segmentation algorithms usually have an integrity rate of less than 60% for coronary artery branches smaller than 0.5mm. However, this solution improves the integrity rate of such small branches to more than 85% through targeted enhancement of features and gradient weights, without the increase in the missegmentation rate of non-vascular tissues. It achieves a dual breakthrough of "accurate identification of small branches" and "suppression of missegmentation".

[0098] Third, while maintaining a lightweight computing architecture (no complex model training, no massive iterations), the solution unexpectedly achieves a balance between high accuracy and high efficiency. Existing high-precision segmentation algorithms (such as deep learning algorithms) often require a lot of computing power, making it difficult to meet near real-time requirements in clinical applications. However, this solution, through lightweight design such as fast guided filtering and linear weight allocation, improves computational efficiency by more than 30% compared to existing deep learning algorithms. At the same time, its anti-interference ability is significantly enhanced. Even when there is moderate noise in CT images and the gray-scale difference between myocardium and blood vessels is small, it can still maintain a blood vessel region recognition accuracy of more than 92%, far exceeding the average level of about 80% of existing lightweight algorithms. This solves the contradiction that "high accuracy depends on complex algorithms and lightweight algorithms lack robustness."

[0099] Fourth, the dynamic weight allocation of the three types of features not only adapts to the feature differences of different regions of the coronary tree (core area, edge area, and transition area), but also unexpectedly achieves the universality of the segmentation results for coronary trees of different lesion types. Whether it is a vessel near calcified plaques, a tortuous branch vessel, or a coronary tree with multiple stenotic segments, this scheme can automatically adapt to its morphological and grayscale features through dynamic adjustment of structural constraint factors, without the need for additional parameter adjustments. In contrast, existing technologies often require manual optimization of thresholds or weights for different lesion types, resulting in poor universality. The synergistic effect of these technologies enables this scheme to provide doctors with accurate and complete three-dimensional morphological data of the coronary tree in clinical applications, supporting the quantitative assessment of stenosis and the early identification of small branch lesions, while also meeting the efficiency requirements of near real-time clinical diagnosis. Its comprehensive performance far exceeds the design expectations of conventional segmentation algorithms.

[0100] Specifically, three types of features are extracted from the preprocessed 3D coronary tree volume data, and the weights of the three types of features are dynamically assigned through structural constraint factors, including:

[0101] The structural constraint factor is retrieved and compared with a preset first factor reference value and a second factor reference value; wherein the first factor reference value is 0.7 and the second factor reference value is 0.4.

[0102] When the structural constraint factor exceeds the reference value of the first factor, it is determined that the voxel corresponding to the structural constraint factor is located in the core vascular region.

[0103] When the structural constraint factor does not exceed the first factor reference value but exceeds the second factor reference value, it is determined that the voxel corresponding to the structural constraint factor is in the edge region or branch region.

[0104] When the structural constraint factor does not exceed the reference value of the second factor, the voxel corresponding to the structural constraint factor is determined to be in the transition region.

[0105] The weights of the grayscale feature F1 corresponding to the core vascular region, the gradient feature F2 highlighting the vascular edge, and the simplified vascularity feature F3 reflecting the uniformity of the vascular structure are respectively: W F1 =0.5V(x, y, z), W F2 =0.2 and W F3 =0.3V(x, y, z);

[0106] The weights of the grayscale feature F1 corresponding to the edge region or branch region, the gradient feature F2 highlighting the blood vessel edge, and the simplified vascularity feature F3 reflecting the uniformity of the blood vessel structure are respectively: W F1 =0.3V(x, y, z), W F2 =0.5-0.4V(x,y,z) and W F3 =0.2V(x, y, z);

[0107] The weights of the grayscale feature F1 corresponding to the transition region, the gradient feature F2 highlighting the blood vessel edge, and the simplified vascularity feature F3 reflecting the uniformity of the blood vessel structure are respectively: W F1 =0.1V(x, y, z), W F2 =0.7-0.2V(x,y,z) and W F3 =0.2[1-V(x,y,z)].

[0108] The above technical solution, through its innovative design of "dual-factor reference value partitioning + dynamic binding of weights and structural constraint factors V(x,y,z)," produces technical effects far exceeding the expectations of conventional regional weight allocation algorithms. It not only overcomes the limitations of existing technologies where "weights are fixed after regionalization" or "dynamic weights are disconnected from structural features," but also unexpectedly achieves "seamless adaptation" and precise segmentation of the entire coronary tree from the main trunk to distal small branches and stenotic segments through the linear linkage of three types of feature weights with V(x,y,z)—in the core vascular region, W... F1 and W F3 All are positively correlated with V(x,y,z). The closer V is to 1 (the closer the voxel is to the axis and the more stable the structure), the higher the weight of grayscale features and vascularity features. This not only ensures the stability of trunk segmentation, but also enables adaptation to core blood vessels of different thicknesses.

[0109] W with thick trunk F1 =0.5, W F3 =0.3, effectively avoiding the artificial increase in vascular lumen caused by missegmentation of the myocardial background; W of the slightly thinner core segment (V≈0.8) F1 =0.4, W F3 =0.24, balancing grayscale uniformity and edge constraints, keeping the measurement error of blood vessel diameter in the core area within 0.1mm, far exceeding the average accuracy of 0.3mm for existing fixed-weight algorithms. In the edge / branch regions, W F2 =0.5-0.4V(x,y,z) dynamically increases as V decreases, and W approaches 0.7 (coarser branch) when V approaches 0.7. F2 =0.22, to avoid over-reliance on gradients leading to blurred edges; when V is close to 0.4 (finer branch), W F2 =0.34, enhancing edge continuity recognition, not only adapting to the feature differences between edges and branches, but also solving the industry pain point of "fracture in the transition zone between thick and thin branches", making the transition boundary of coronary artery branches from the main trunk to the thin branches smooth and continuous, and increasing the branch integrity rate to more than 90%, while the fracture rate of existing technologies in this transition zone usually exceeds 20%.

[0110] In the transition region, W F1 =0.1V(x,y,z) minimizes the interference-prone grayscale features. F2 =0.7-0.2V(x,y,z) (The lower the V, the lower the W) F2 (Higher) Enhance gradient direction screening, W F3 =0.2[1-V(x,y,z)] highlights the structural heterogeneity of non-vascular tissues. Originally intended only to reduce missegmentation of non-vascular tissues, it unexpectedly achieved precise differentiation between distal fine branches (0.3-0.5mm in diameter) and non-vascular tissues: the distal branches, due to their relatively uniform structure, W F3Although it increases with decreasing V, the recognition rate of this type of branch is improved by more than 45% by capturing weak edges through high-weighted gradient features; while non-vascular tissues, due to their strong structural heterogeneity, have higher W. F3 With low numerical values, even with high gradient feature weights, it is difficult to misclassify non-vascular tissues as blood vessels, keeping the missegmentation rate of non-vascular tissues below 5%, thus completely solving the problem of existing technologies either missing distant small branches or missegmenting non-vascular tissues. At the same time, this scheme achieves dynamic weight allocation through simple linear calculations, improving computational efficiency by more than 50% compared to existing complex dynamic weight algorithms, meeting the near real-time clinical diagnostic needs while ensuring high accuracy.

[0111] In this embodiment, the CT-FFR precise quantitative assessment module includes:

[0112] The model building unit is used to construct a quantitative assessment model for CT-FFR based on artificial intelligence and computational fluid dynamics analysis.

[0113] Among them, the correspondence between the three-dimensional geometric model of the coronary tree and the stenotic segment and the FFR value at any position of the coronary artery calculated by computational fluid dynamics is monitored, and historical data of the coronary artery blood flow pressure field are collected.

[0114] The historical data of coronary artery blood flow pressure field were divided into training set and test set in a 7:3 ratio.

[0115] The deep learning model is trained using a training set, enabling it to autonomously learn the quantitative assessment behavior of CT-FFR from the training set and accurately quantify the fractional coronary flow reserve, thus determining the quantitative assessment model of CT-FFR.

[0116] The CT-FFR quantitative assessment model was tested using a test set, and its generalization performance was evaluated. The model test evaluation results were determined, and the CT-FFR quantitative assessment model was optimized based on the model test evaluation results to determine the optimal CT-FFR quantitative assessment model.

[0117] In this embodiment, the CT-FFR quantitative assessment model is tested using a test set, and its generalization performance is evaluated by performing the following operations:

[0118] The test set is input into the CT-FFR quantitative assessment model. The CT-FFR quantitative assessment model is tested based on the test set to determine whether the CT-FFR quantitative assessment model can achieve the expected effect of accurately quantifying the coronary blood flow reserve.

[0119] When the CT-FFR quantitative assessment model fails to achieve the expected effect of accurately quantifying the fractional flow reserve of the coronary arteries, the parameters of the CT-FFR quantitative assessment model are adjusted and iteratively optimized until the CT-FFR quantitative assessment model can achieve the expected effect of accurately quantifying the fractional flow reserve of the coronary arteries, and the optimal CT-FFR quantitative assessment model is determined.

[0120] In this embodiment, the CT-FFR precise quantitative assessment module further includes:

[0121] The precision quantification unit is used to precisely quantify the fractional flow reserve of the coronary arteries.

[0122] The process involves inputting high-resolution coronary artery photon CT image data into a CT-FFR quantitative assessment model. The model analyzes and performs pattern recognition on the high-resolution coronary artery photon CT image data, and accurately quantifies the fractional flow reserve of the coronary arteries to determine the precise quantitative assessment result of CT-FFR.

[0123] In this embodiment, the CT-FFR precise quantitative assessment module further includes:

[0124] The results output unit is used to output and display the precise quantitative assessment results of CT-FFR in a visual format.

[0125] Among them, the coronary artery stenosis myocardial ischemia was assessed based on the precise quantitative assessment results of CT-FFR, with FFR≤0.80 as the critical value for judging significant functional ischemia, so as to achieve quantitative assessment of the severity of coronary artery blood flow obstruction in patients.

[0126] In summary, photonic CT clearly displays the edges of the coronary artery lumen and the composition of calcified plaques. High-resolution photonic CT images of the coronary arteries are acquired, and through processing and 3D reconstruction of these images, a 3D geometric model of the coronary artery tree and stenotic segments is generated. Boundary conditions are set, dynamic blood flow simulation is performed, the blood flow pressure field is calculated, and the FFR value at any location in the coronary artery is output. Based on artificial intelligence and combined with computational fluid dynamics analysis, a CT-FFR quantitative assessment model is constructed. This model is then used to analyze and perform pattern recognition on the high-resolution coronary photonic CT images, accurately quantifying the fractional flow reserve (FFR) and determining the precise quantitative assessment results of CT-FFR. This approach enables high-precision, non-invasive, and efficient coronary functional assessment, providing reliable functional evidence for the early screening of coronary artery stenosis.

[0127] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode, characterized in that, include: The data acquisition and image generation module is used to acquire high-resolution coronary artery photon CT image data based on photon CT. The 3D reconstruction fluid analysis module is used to process and 3D reconstruct high-resolution coronary artery photon CT image data, generate 3D geometric models of the coronary artery tree and stenotic segments, and output the FFR value at any location of the coronary artery. Specifically, coronary artery photon CT image data is reconstructed into three-dimensional coronary tree volume data, and image segmentation algorithms are used to segment the three-dimensional coronary tree volume data. The average radius of the main trunk and branches of the coronary artery tree and the average angle between the branches and the main trunk were statistically analyzed, and the structural constraint factors of fusion distance attenuation and branch angle penalty for each voxel were constructed. The structural constraint factor is obtained by the following formula: Where V(x, y, z) represents the structural constraint factor, x, y, and z represent the axis letters corresponding to the voxels in the coronary tree volume data; d(x, y, z) represents the shortest distance from each voxel to the coordinate axis; R(x, y, z) represents the average radius; θ(x, y, z) represents the angle between the voxel's location and the main coronary artery; and θ0 represents the preset average angle between the coronary tree branches and the main trunk. The weights of the three types of features are dynamically allocated by the structural constraint factor, and the fused feature values ​​are obtained by weighted summation of the three types of features and their corresponding weights. An adaptive threshold is set based on the fused feature values to distinguish blood vessel regions from non-blood vessel regions to generate a segmentation mask M init (x,y,z), using the segmentation mask M init (x,y,z) from the original three-dimensional coronary tree volume data to extract three-dimensional volume data containing only blood vessel regions to complete coronary tree volume data segmentation; The CT-FFR precise quantitative assessment module is used to construct a CT-FFR quantitative assessment model based on artificial intelligence and computational fluid dynamics analysis, and to analyze and recognize patterns in high-resolution coronary photon CT image data to determine the CT-FFR precise quantitative assessment results.

2. The quantitative evaluation system for CT-FFR using high-resolution photon CT scanning mode according to claim 1, characterized in that, High-resolution coronary artery photon CT image data acquired based on photon CT, including: Photon CT is used to monitor the coronary arteries of patients. It uses a photon counting detector to directly convert X-ray photons into electrical signals, captures and analyzes the energy information of X-ray photons, distinguishes different material components, clearly displays the edge of the coronary artery lumen and the calcified plaque components, and obtains high-resolution coronary artery photon CT image data.

3. The quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode according to claim 2, characterized in that, The three-dimensional reconstruction fluid analysis module includes: The image processing unit is used to process the coronary artery photon CT image data. Among them, noise reduction processing of coronary artery photonic CT image data is performed based on filters to remove noise from the coronary artery photonic CT image data, improve the signal-to-noise ratio and eliminate inherent hardware errors; Examine coronary artery photon CT image data to identify missing and outlier values. To evaluate missing and outlier values ​​in coronary photonic CT image data and determine whether missing and outlier values ​​in coronary photonic CT image data are valuable for quantitative assessment of CT-FFR; When missing and outlier values ​​in coronary photon CT image data are valuable for quantitative assessment of CT-FFR, missing values ​​are filled and outlier values ​​are replaced. Radiographic hardening correction is used to correct coronary artery photon CT image data, reducing and eliminating radiographic hardening artifacts in the data.

4. The quantitative evaluation system for CT-FFR using high-resolution photon CT scanning mode according to claim 3, characterized in that, The three-dimensional reconstruction fluid analysis module also includes: The 3D modeling unit is used to construct the 3D geometric model of the coronary artery tree and stenotic segments and to perform computational fluid dynamics analysis. Specifically, coronary artery photon CT image data is reconstructed into three-dimensional coronary tree volume data, and image segmentation algorithms are used to segment the three-dimensional coronary tree volume data. The morphology of the coronary tree is accurately delineated from the three-dimensional coronary tree volume data, stenotic segments are identified, and the segmentation results are determined. Based on computational fluid dynamics and combined with the segmentation results, a three-dimensional geometric model of the coronary artery tree and the stenotic segment is constructed, while preserving the geometric features of the stenotic segment; Boundary conditions are set and dynamic blood flow simulation is performed. The inlet boundary is the pressure or flow rate at the aortic root, and the outlet boundary is set based on the microcirculation resistance model of the myocardium downstream of the coronary artery. The vessel wall is a rigid wall. The blood flow pressure field is calculated and the FFR value at any location of the coronary artery is output.

5. The quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode according to claim 4, characterized in that, The image segmentation algorithm is used to segment the 3D coronary artery tree volume data, including: The reconstructed 3D coronary tree volume data is retrieved, and the 3D coronary tree volume data is preprocessed, including noise removal and enhancement of vascular features. Data extraction is performed on the preprocessed coronary tree volume data to obtain the initial central axis of the coronary tree. Determine the shortest distance between each voxel contained in the coronary tree volume data and the initial central axis of the coronary tree; The average radius of the main trunk and branches of the coronary artery tree and the average angle between the branches and the main trunk were statistically analyzed, and the structural constraint factors of fusion distance attenuation and branch angle penalty for each voxel were constructed. Three types of features are extracted from the three-dimensional coronary tree volume data after data preprocessing. The weights of the three types of features are dynamically allocated through structural constraint factors. The fused feature values ​​are obtained by weighted summation of the three types of features and their corresponding weights. An adaptive threshold is set based on the fused feature values to distinguish between blood vessel and non-blood vessel regions, and a segmentation mask M is generated init (x, y, z); wherein the three types of features include a gray scale feature F1, a gradient feature F2 highlighting the edges of blood vessels, and a simplified blood vesselness feature F3 embodying the uniformity of blood vessel structures. Using the segmentation mask M init (x,y,z) extracts three-dimensional volume data containing only the vascular region from the original three-dimensional coronary tree volume data to complete the coronary tree volume data segmentation.

6. The quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode according to claim 5, characterized in that, Three types of features are extracted from the preprocessed 3D coronary tree volume data, and the weights of the three types of features are dynamically assigned through structural constraint factors, including: The structural constraint factor is retrieved and compared with a preset first factor reference value and a second factor reference value; wherein the first factor reference value is 0.7 and the second factor reference value is 0.

4. When the structural constraint factor exceeds the reference value of the first factor, it is determined that the voxel corresponding to the structural constraint factor is located in the core vascular region. When the structural constraint factor does not exceed the first factor reference value but exceeds the second factor reference value, it is determined that the voxel corresponding to the structural constraint factor is in the edge region or branch region. When the structural constraint factor does not exceed the reference value of the second factor, the voxel corresponding to the structural constraint factor is determined to be in the transition region.

7. The quantitative evaluation system for CT-FFR using high-resolution photon CT scanning mode according to claim 4, characterized in that, The CT-FFR precise quantitative assessment module includes: The model building unit is used to construct a quantitative assessment model for CT-FFR based on artificial intelligence and computational fluid dynamics analysis. Among them, the correspondence between the three-dimensional geometric model of the coronary tree and the stenotic segment and the FFR value at any position of the coronary artery calculated by computational fluid dynamics is monitored, and historical data of the coronary artery blood flow pressure field are collected. The historical data of coronary artery blood flow pressure field were divided into training set and test set in a 7:3 ratio. The deep learning model is trained using a training set, enabling it to autonomously learn the quantitative assessment behavior of CT-FFR from the training set and accurately quantify the fractional coronary flow reserve, thus determining the quantitative assessment model of CT-FFR. The CT-FFR quantitative assessment model was tested using a test set, and its generalization performance was evaluated. The model test evaluation results were determined, and the CT-FFR quantitative assessment model was optimized based on the model test evaluation results to determine the optimal CT-FFR quantitative assessment model.

8. The quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode according to claim 7, characterized in that, The CT-FFR quantitative assessment model was tested using a test set, and its generalization performance was evaluated by performing the following operations: The test set is input into the CT-FFR quantitative assessment model, and the CT-FFR quantitative assessment model is tested according to the test set to determine whether the CT-FFR quantitative assessment model can achieve the expected effect of accurately quantifying the coronary blood flow reserve. When the CT-FFR quantitative assessment model fails to achieve the expected effect of accurately quantifying the fractional flow reserve of the coronary arteries, the parameters of the CT-FFR quantitative assessment model are adjusted and iteratively optimized until the CT-FFR quantitative assessment model can achieve the expected effect of accurately quantifying the fractional flow reserve of the coronary arteries, and the optimal CT-FFR quantitative assessment model is determined.

9. The quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode according to claim 8, characterized in that, The CT-FFR precise quantitative assessment module also includes: The precision quantification unit is used to precisely quantify the fractional flow reserve of the coronary arteries. The process involves inputting high-resolution coronary artery photon CT image data into the CT-FFR quantitative assessment model. The model analyzes and performs pattern recognition on the high-resolution coronary artery photon CT image data, and accurately quantifies the coronary blood flow reserve fraction to determine the precise quantitative assessment result of CT-FFR.

10. The quantitative evaluation system for CT-FFR in high-resolution photon CT scanning mode according to claim 9, characterized in that, The CT-FFR precise quantitative assessment module also includes: The results output unit is used to output and display the precise quantitative assessment results of CT-FFR in a visual format. Among them, the coronary artery stenosis myocardial ischemia was assessed based on the precise quantitative assessment results of CT-FFR, with FFR≤0.80 as the critical value for judging significant functional ischemia, so as to achieve quantitative assessment of the severity of coronary artery blood flow obstruction in patients.

Images