Cerebrovascular stenosis three-dimensional reconstruction method based on multi-modal image fusion

Abstract

The invention relates to the technical field of image reconstruction, and provides a cerebrovascular stenosis three-dimensional reconstruction method based on multi-modal image fusion, and the method comprises the steps: collecting multi-modal image data of a patient, including an arterial phase CTA sequence and a 3D TOF-MRA sequence, and obtaining a plurality of voxels; obtaining the modal contradiction degree of each voxel; screening a plurality of preliminary calcification voxels according to the CTA data of the voxels, and obtaining calcification influence factors of the voxels in combination with the influence range of the flowering artifacts; acquiring CTA local signal reliability of each voxel; acquiring local signal unevenness of each voxel; determining a slow blood flow index of each voxel; obtaining the MRA signal reliability of each voxel; and weight distribution is carried out on the two-modal data and registration fusion is carried out. The method aims at solving the problem that the quality of a reconstructed image is reduced due to the fact that multi-modal data is changed by hemodynamics under the situation of cerebrovascular stenosis.

Description

Technical Field

[0001] This invention relates to the field of image reconstruction technology, and more specifically to a three-dimensional reconstruction method for cerebral vascular stenosis based on multimodal image fusion. Background Technology

[0002] Cerebral vascular stenosis is one of the main causes of ischemic stroke. Accurate assessment of the degree of stenosis, plaque morphology, and hemodynamic characteristics is crucial for clinical diagnosis, treatment planning, and prognostic evaluation. Currently, clinical assessment mainly relies on two non-invasive imaging techniques: computed tomography angiography (CTA) and magnetic resonance angiography (MRA). CTA has high spatial resolution and calcification identification capabilities, but it is susceptible to calcification "flowering artifacts," which may overestimate the degree of stenosis in areas of severe calcification. MRA is sensitive to slow blood flow and is unaffected by calcification, but it is prone to signal loss due to turbulence, which may underestimate the degree of stenosis. Single-modal imaging has inherent limitations and cannot comprehensively and accurately reflect the complex pathological state of cerebrovascular diseases.

[0003] In existing technologies, three-dimensional reconstruction of cerebral vascular stenosis is performed by superimposing images of different modalities. However, the characteristics of different modal data are not taken into account, and it is difficult to optimize for the specific scenario of cerebral vascular stenosis. In particular, the reconstruction effect is not ideal in the stenotic area due to changes in hemodynamics and image quality degradation. CTA has high resolution but poor soft tissue contrast, while MRA is sensitive to slow blood flow but has low spatial resolution. In specific cerebral vascular stenosis areas, CTA may produce artifacts due to calcified plaques, while MRA may lose signals due to slow blood flow, making it difficult to obtain effective three-dimensional reconstruction results. Summary of the Invention

[0004] This invention provides a three-dimensional reconstruction method for cerebral vascular stenosis based on multimodal image fusion, to solve the problem of degraded reconstructed image quality caused by hemodynamic changes in cerebral vascular stenosis scenarios. The specific technical solution adopted is as follows: This invention proposes a three-dimensional reconstruction method for cerebral vascular stenosis based on multimodal image fusion, which includes the following steps: Acquire multimodal imaging data of the patient, including arterial phase CTA sequences and 3D TOF-MRA sequences, and obtain several voxels; Based on the difference in the probability of the same voxel representing blood vessels in different modal data, the modal contradiction degree of each voxel is obtained; several preliminary calcified voxels are screened based on the voxel CTA data, and the calcification influence factor of each voxel is obtained by combining the influence range of flowering artifact; considering the distribution relationship between the voxel CTA data and the vascular CT value performance, the CT value reliability of each voxel is obtained; and by combining the calcification influence factor and the CT value reliability, the CTA local signal reliability of each voxel is obtained. Based on the differences in MRA signal intensity within the vicinity of a voxel, the local signal inhomogeneity of each voxel is obtained; the vascular layer of each voxel is obtained through morphological analysis, and the signal intensity of different voxels at the same vascular layer is analyzed. Combined with the distribution of vascular layers, the slow blood flow index of each voxel is determined; the reliability of the MRA signal of each voxel is obtained by combining the local signal inhomogeneity and the slow blood flow index. Based on the reliability of the CTA local signal and the reliability of the MRA signal, as well as the degree of modal inconsistency of each voxel, weights are assigned to the two modal data and registration and fusion are performed.

[0005] Optionally, the modal contradiction degree of each voxel is obtained by the following method: The CTA and MRA data were standardized separately, and the converted data represented the probability of the representative vessels of each of the two modalities in each voxel. For any voxel, the absolute value of the difference between the two modal data and the representative blood vessel probability of that voxel is taken as the degree of modal contradiction of that voxel.

[0006] Optionally, the specific method for obtaining the calcification influencing factors of each voxel includes: For any voxel, obtain the Euclidean distance between the voxel and its nearest pre-calcified voxel. Use the ratio of the Euclidean distance to the defined radius of influence as the calcification distance factor of the voxel. Subtract the calcification distance factor of the voxel from 1 and output the result through a linear rectified function as the calcification influence factor of the voxel.

[0007] Optionally, the specific method for obtaining the confidence level of the CT values ​​of each voxel includes: The CTA data of each voxel are used as the CT value of each voxel; preset calcification threshold, initial vascular threshold and ideal vascular threshold are set. If the CT value of any voxel is less than or equal to the initial vascular threshold, or greater than or equal to the calcification threshold, the confidence level of the CT value of that voxel is set to 0. If the CT value of the voxel is greater than the initial vascular threshold and less than or equal to the ideal vascular threshold, the difference between the CT value of the voxel and the initial vascular threshold is taken as the first difference, the difference between the ideal vascular threshold and the initial vascular threshold is taken as the first standard deviation, and the ratio of the first difference to the first standard deviation is taken as the confidence level of the CT value of the voxel. If the CT value of the voxel is greater than the ideal vascular threshold and less than the calcification threshold, the difference between the CT value of the voxel and the calcification threshold is taken as the second difference. The difference between the ideal vascular threshold and the calcification threshold is taken as the second standard deviation. The ratio of the second difference to the second standard deviation is taken as the confidence level of the CT value of the voxel.

[0008] Optionally, the CTA local signal reliability of each voxel is obtained using the following method: The difference obtained by subtracting the calcification influence factor of any voxel from 1, multiplied by the product of the CT value confidence of that voxel, is used as the CTA local signal reliability of that voxel.

[0009] Optionally, the local signal non-uniformity of each voxel is obtained by the following method: The signal intensity corresponding to each voxel in the 3D TOF-MRA data is used as the MRA signal intensity of each voxel; the initial vascular mask is obtained; for any voxel, a spatial neighborhood of the voxel is constructed with the voxel as the center and the voxel in a range of 3×3×3, and the spatial neighborhood contains the voxel. The coefficient of variation of the MRA signal intensity of voxels belonging to the initial vascular mask in the spatial neighborhood of the voxel is used as the local signal inhomogeneity of the voxel.

[0010] Optionally, the specific method for obtaining the vascular layers of each voxel through morphological analysis includes: By refining the morphology, a rough vascular centerline skeleton is extracted. For any voxel, based on the vascular centerline skeleton of the nearest voxel to it, the voxel is taken as the voxel of that segment of the blood vessel. The vascular tree structure is obtained by segmenting and tracing blood vessels. The branches of the vascular tree structure are numbered from small to large along the branch direction starting from the main trunk of the blood vessel, and each segment of the blood vessel is assigned a corresponding level. Thus, the level of the blood vessel corresponding to the voxel is obtained, which is used as the vascular level of the voxel.

[0011] Optionally, the slow blood flow index of each voxel is obtained using the following method: For any voxel, obtain the mean MRA signal intensity of all voxels in the vascular layer where the voxel is located, and use the ratio of the voxel's MRA signal intensity to the mean MRA signal intensity as the signal deviation factor of the voxel; multiply the difference obtained by subtracting the signal deviation factor from 1 by the product of the ratio of the voxel's vascular layer to the maximum value among all vascular layers, and use the product as the slow flow index of the voxel.

[0012] Optionally, the specific method for assigning weights and performing registration and fusion of the two modal data based on the reliability of the CTA local signal and the reliability of the MRA signal, as well as the degree of modal inconsistency of each voxel, includes: For any voxel, the modal data corresponding to the maximum value between the CTA local signal reliability and the MRA signal reliability of that voxel is used as the reference modal data of that voxel. Obtain the sum of the CTA local signal reliability and MRA signal reliability of the voxel. Use the ratio of the signal reliability corresponding to the reference modal data to the sum as the initial reference modal weight of the voxel. Multiply the sum obtained by adding 1 to the modal inconsistency degree of the voxel by the initial reference modal weight as the reference modal weight of the voxel and assign it to the corresponding modal data. Based on the two-modal data of each voxel and their corresponding weights, a fused 3D volume is obtained.

[0013] Optionally, the specific method for fusing the two-modal data of each voxel and their corresponding weights to obtain the fused 3D volume includes: For any voxel, based on the modal weights corresponding to the two modal data of that voxel, the representative vessel probabilities after the two modal data of that voxel are standardized are weighted and fused to obtain the fused vessel probability of that voxel. The fusion vessel probability of each voxel is obtained to get new three-dimensional volume data; the new three-dimensional volume data is then subjected to three-dimensional smoothing to obtain the fused three-dimensional volume.

[0014] The beneficial effects of this invention are as follows: This invention, through collaborative acquisition and standardized preprocessing of multimodal image data, extracts multidimensional physical features reflecting calcification effects, loss of turbulent signals, and slow blood flow sensitivity from aligned CTA and MRA data, respectively. It also extracts the turbulence index based on local signal statistics and the slow blood flow index based on vessel tracking topology and relative signal intensity. Subsequently, by comparing the vessel probabilities of CTA and MRA, conflicting regions are identified. Based on the conflict type and the deep features of the corresponding region, physical constraint rules are used to dynamically allocate modal fusion weights for each voxel. These weights guide the adaptive fusion of each voxel, generating conflict-mitigated fused three-dimensional volume data, thereby achieving three-dimensional reconstruction of cerebral vascular stenosis through multimodal image fusion. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of a three-dimensional reconstruction method for cerebral vascular stenosis based on multimodal image fusion provided in an embodiment of the present invention. Figure 2 Example graphs of normalized CTA and MRA data; Figure 3 This is a diagram illustrating the weighting adjustment of CTA data. Figure 4 This is a schematic diagram illustrating the fusion of MRA and CTA data. Detailed Implementation

[0017] 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.

[0018] Please see Figure 1 The diagram illustrates a flowchart of a three-dimensional reconstruction method for cerebral vascular stenosis based on multimodal image fusion according to an embodiment of the present invention. The method includes the following steps: Step S001: Acquire the patient's multimodal imaging data, including arterial phase CTA sequences and 3D TOF-MRA sequences, and obtain several voxels.

[0019] The purpose of this embodiment is to achieve three-dimensional reconstruction by fusing multimodal image data of patients with cerebral vascular stenosis. Therefore, it is necessary to first collect multimodal image data of the patients.

[0020] Specifically, based on clinical standard scanning protocols, multimodal imaging data of the patient's cerebral blood vessels are acquired simultaneously while maintaining the patient's position and physiological state. The patient's arterial phase CTA sequence is obtained through computed tomography angiography, and the patient's three-dimensional time-of-flight magnetic resonance angiography sequence, i.e., 3D TOF-MRA sequence, is obtained through magnetic resonance angiography. At the same time, the patient's plain CT sequence is acquired for subsequent calcification identification. This ensures that all data are comparable in terms of spatial coverage and anatomical location, and all data are exported in DICOM format for subsequent processing.

[0021] Furthermore, the multimodal data are aligned to the same three-dimensional coordinate system through rigid or non-rigid registration, and several voxels are labeled in the three-dimensional coordinate system.

[0022] Step S002: Based on the difference in the probability of the same voxel representing blood vessels in different modal data, obtain the modal contradiction degree of each voxel; based on the CTA data of the voxels, screen several preliminary calcified voxels, and combine the influence range of flowering artifacts to obtain the calcification influence factor of each voxel; consider the distribution relationship between the CTA data of voxels and the CT value of blood vessels to obtain the CT value reliability of each voxel, and then obtain the CTA local signal reliability of each voxel.

[0023] It should be noted that by analyzing the characteristics of narrow regions and dynamically selecting data sources, more effective three-dimensional reconstruction results can be obtained.

[0024] Preferably, in one embodiment of the present invention, the modal inconsistency degree of each voxel is obtained based on the difference in the probability of the same voxel representing blood vessels in different modal data, including the following specific method: CTA and MRA data were standardized (maximum and minimum value normalization) to ensure that their values ​​across all voxels were between 0 and 1, eliminating dimensional differences. After transformation, each data point represents the probability of a representative vessel for each modality in each voxel. For any voxel, the absolute value of the difference between the probabilities of the two modalities representing the representative vessels in that voxel was used as the degree of modal contradiction for that voxel. Figure 2 As shown, it illustrates the normalization results of CTA and MRA data.

[0025] It should be noted that the degree of modal inconsistency reflects the consistency of the two modal imaging judgments at this voxel. The larger the inconsistency index, the more inconsistent the judgments of the two modalities on the probability of blood vessels at this voxel. Further analysis of the root cause of the inconsistency is needed to determine the final analysis weight of this voxel and obtain an effective three-dimensional reconstruction result.

[0026] It should be further noted that in narrow regions, CTA and MRA have different error patterns. By analyzing the local features around each voxel, it is possible to determine which modality is more reliable, thereby assigning weights and effectively reconstructing the patient's cerebral blood vessels in three dimensions. CTA overestimates stenosis due to calcified plaque blooming artifacts, while MRA underestimates stenosis due to the loss of turbulent signals. Simple weighted averaging leads to systematic errors. Therefore, for multimodal data, it is necessary to quantify the reliability of each data separately and then achieve weighted fusion to obtain effective three-dimensional reconstruction results.

[0027] Preferably, in one embodiment of the present invention, several preliminary calcified voxels are screened based on the CTA data of the voxels, and the calcification influencing factors of each voxel are obtained by combining the influence range of flowering artifacts. The specific method includes: It should be noted that in CTA, the vascular lumen should show a high CT value (due to the contrast agent). If a voxel has a high CT value in CTA but there is calcification around it (which may produce flowering artifacts), that is, the high signal in the calcified area will spread to the surrounding tissue, causing the CT value of the adjacent area to be abnormally increased. This artifact will blur the boundary of the vascular lumen and may even cause the lumen to be incorrectly enlarged, thus overestimating the degree of stenosis, thereby reducing its reliability. The high CT value may be due to the actual contrast agent filling of the vascular lumen or it may be due to the artifact diffusion of calcified plaques. Therefore, the reliability of voxel CTA needs to be further quantified.

[0028] Specifically, in the plain CT sequence, the plain CT value of each voxel is obtained, and a calcification threshold is preset. In this embodiment, the calcification threshold is described as 350 HU. If the plain CT threshold of any voxel is greater than the calcification threshold, the voxel is regarded as a preliminary calcified voxel. For any preliminary calcified voxel, the influence range of the flowering artifact is defined. In this embodiment, the influence radius is defined as 3 mm. Then, the spherical region formed by the preliminary calcified voxel as the center and the influence radius is regarded as the influence range of the flowering artifact. The unit size of the voxel is determined according to the existing algorithm.

[0029] Furthermore, for any voxel, the Euclidean distance between the voxel and its nearest pre-calcified voxel is obtained. The ratio of this Euclidean distance to the influence radius is used as the calcification distance factor for that voxel. The difference between 1 and the calcification distance factor is then processed through a linear rectified function and used as the calcification influence factor for that voxel. The linear rectified function is... This function ensures the non-negativity of the result; if the input is negative, it outputs 0; otherwise, it outputs the result directly.

[0030] It should be noted that the smaller the calcification distance factor, the greater the influence of the flowering artifact of the initially calcified voxel on the voxel, and the greater its calcification influence factor.

[0031] It should be further noted that under the influence of strong artifacts, calcification artifacts may cause abnormally high CT values. If the CT value is particularly high (e.g., exceeding the calcification threshold), then the voxel itself may be a calcified voxel, further reducing the reliability.

[0032] Preferably, in one embodiment of the present invention, considering the distribution relationship between voxel CTA data and vascular CT values, the reliability of CT values ​​for each voxel is obtained, thereby acquiring the reliability of local CTA signals for each voxel. The specific method includes: The CTA data of each voxel are used as the CT value of each voxel. A calcification threshold, an initial vascular threshold, and an ideal vascular threshold are preset. In this embodiment, the calcification threshold is described as 350 HU, and the initial vascular threshold and ideal vascular threshold are described as 150 HU and 270 HU, respectively. If the CT value of any voxel is less than or equal to the initial vascular threshold, or greater than or equal to the calcification threshold, the confidence level of the voxel's CT value is set to 0, meaning that the voxel is either not a vascular voxel or is a calcified voxel. If the CT value of the voxel is greater than the initial vascular threshold and less than or equal to the ideal vascular threshold, the confidence level of the voxel's CT value increases with the increase of the CT value, and the CT value of the voxel is subtracted. The difference obtained from the initial vascular threshold is used as the first difference. The difference obtained by subtracting the initial vascular threshold from the ideal vascular threshold is used as the first standard deviation. The ratio of the first difference to the first standard deviation is used as the confidence level of the CT value of the voxel. If the CT value of the voxel is greater than the ideal vascular threshold and less than the calcification threshold, the confidence level of the CT value of the voxel decreases as the CT value increases. The difference obtained by subtracting the CT value of the voxel from the calcification threshold is used as the second difference. The difference obtained by subtracting the ideal vascular threshold from the calcification threshold is used as the second standard deviation. The ratio of the second difference to the second standard deviation is used as the confidence level of the CT value of the voxel.

[0033] It should be further explained that, considering the spatial influence of calcification artifacts, which can spread to surrounding tissues, a calcification influence factor is introduced. This factor is based on the distance from the voxel to the nearest calcified voxel. The closer the distance, the greater the influence of calcification artifacts. Even if the CT value has high reliability, its reliability will decrease if it is greatly affected by calcification artifacts. Conversely, if the CT value has low reliability, its reliability will also be low even if it is not affected by calcification.

[0034] Furthermore, the difference obtained by subtracting the calcification influence factor of any voxel from 1 is multiplied by the product obtained by the CT value confidence of that voxel, and this product is used as the CTA local signal reliability of that voxel; the CTA local signal reliability of each voxel is obtained according to the above method.

[0035] Thus, the reliability of the local CTA signal for each voxel is obtained.

[0036] Step S003: Based on the differences in MRA signal intensity within the vicinity of a voxel, obtain the local signal inhomogeneity of each voxel; obtain the vascular layer of each voxel through morphological analysis, analyze the signal intensity of different voxels at the same vascular layer, and determine the slow blood flow index of each voxel by combining the vascular layer distribution; and obtain the MRA signal reliability of each voxel by combining the local signal inhomogeneity and the slow blood flow index.

[0037] It should be noted that in MRA, each voxel represents the magnetic resonance signal intensity. Within the blood vessel lumen, flowing blood produces a high signal, while static tissue has a lower signal. The blood vessel lumen should show a high signal, but in the turbulent region after stenosis, the signal may be lost, making the blood vessel appear narrower than it actually is. This is the low point in MRA data that corresponds to the degree of cerebral vascular stenosis. In addition, if the flow velocity at this location is high in phase-contrast MRA, TOF-MRA may reduce the signal due to flow phase loss. In the turbulent region after stenosis, the direction and magnitude of blood flow velocity change drastically, causing phase signal loss. Therefore, turbulence is estimated by the inhomogeneity of local signals.

[0038] Preferably, in one embodiment of the present invention, the local signal non-uniformity of each voxel is obtained based on the difference in MRA signal intensity within the vicinity of the voxel, including the following specific method: The signal intensity corresponding to each voxel in the 3D TOF-MRA data is used as the MRA signal intensity of each voxel. An initial vascular mask is obtained. For any voxel, a spatial neighborhood of voxels with a range of 3×3×3 centered on the voxel is constructed. The spatial neighborhood includes the voxel. If the voxel is close to the image boundary and a complete spatial neighborhood cannot be obtained, the spatial neighborhood is constructed using the actual voxels. The coefficient of variation of the MRA signal intensity of the voxels belonging to the initial vascular mask in the spatial neighborhood of the voxel, i.e., the ratio of the standard deviation to the mean, is used as the local signal non-uniformity of the voxel. In order to avoid the denominator being 0 during the ratio calculation, a hyperparameter is added to the denominator. In this embodiment, the hyperparameter is described as 0.01. If the number of voxels belonging to the initial vascular mask in the spatial neighborhood is less than or equal to 1, the corresponding local signal non-uniformity is set to 0.

[0039] It should be noted that the larger the coefficient of variation, i.e., the larger the ratio, the more severe the loss of turbulent signal at that voxel may be. In other words, the current MRA signal intensity at that voxel may be at a low trough, and thus its reliability is lower.

[0040] It should be further noted that MRA is sensitive to slow blood flow, while CTA may show weak signal in slow blood flow areas due to insufficient contrast agent filling. Therefore, MRA may be more reliable in slow blood flow areas. However, slow blood flow areas may also have low signal and be easily confused with the background. Therefore, it is necessary to quantify the possibility that a voxel belongs to slow blood flow but is still a blood vessel. Using the signal intensity of TOF-MRA and the prior of vascular structure, it can be estimated that in small vessels downstream or distal to stenosis, the blood flow velocity is slow. Therefore, the vascular tree obtained from vascular tracing should show that the vascular region is continuous. The vascular continuity and signal intensity can be used to estimate the slow blood flow status.

[0041] Preferably, in one embodiment of the present invention, the vascular layer of each voxel is obtained through morphological analysis, the signal intensity of different voxels at the same vascular layer is analyzed, and the slow blood flow index of each voxel is determined in combination with the vascular layer distribution. The specific method includes: Morphological refinement is used to extract a rough vascular centerline skeleton. For any voxel, based on the vascular centerline skeleton closest to it, the voxel is taken as the voxel of that segment of the blood vessel. The vascular tree structure is obtained through vascular segmentation and tracing (existing methods, which will not be described in this embodiment). Through the branches of the vascular tree structure, the blood vessel segments are numbered incrementally from the main trunk along the branch direction, and each segment of the blood vessel is assigned a corresponding level. Thus, the level of the blood vessel segment corresponding to the voxel is obtained, which is taken as the vascular level of the voxel. The vascular levels are numbered from the main trunk from the main trunk, with the vascular level of the main trunk being 0.

[0042] It should be further explained that the determination of whether it is slow blood flow is based on the blood vessel layer where the voxel is located (the more distal the layer, the higher the signal intensity) and the signal intensity. It is assumed that the main trunk (layer 0) is fast blood flow with high signal intensity. As the layer increases, the blood flow speed slows down and the signal intensity gradually decreases.

[0043] Furthermore, for any voxel, the mean MRA signal intensity of all voxels in the vascular layer where the voxel is located is obtained, and the ratio of the voxel's MRA signal intensity to the mean MRA signal intensity is used as the signal deviation factor of the voxel; the difference obtained by subtracting the signal deviation factor from 1 is multiplied by the product of the ratio of the voxel's vascular layer to the maximum value among all vascular layers, and this product is used as the slow blood flow index of the voxel; if the signal deviation factor is greater than 1, the slow blood flow index of the voxel is set to 0.

[0044] It should be noted that a signal deviation factor less than 1 and the smaller the value, the lower the signal of the voxel is compared to the average at the same level, which may be due to slow blood flow or noise. However, when the signal intensity is lower than the average level at the same level in distal vessels (larger levels), the slow blood flow index is higher, indicating that the voxel may be in a slow blood flow region. In this case, the reliability of the MRA data is relatively high, meaning that slow blood flow is more common in distal vessels.

[0045] It should be further noted that when the signal is in a distal blood vessel and the signal is lower than the average at the same level, the slow flow sensitivity is high, indicating that the voxel may belong to a slow flow region, and MRA may be more reliable than CTA in this area; at the same time, the smaller the local signal inhomogeneity, the higher the reliability of the corresponding MRA data.

[0046] Preferably, in one embodiment of the present invention, the reliability of the MRA signal of each voxel is obtained by combining the local signal non-uniformity and the slow blood flow index, and the specific method includes: The MRA signal reliability of any voxel is obtained based on the local signal inhomogeneity and slow blood flow index. The MRA signal reliability is negatively correlated with the local signal inhomogeneity and positively correlated with the slow blood flow index. The MRA signal reliability has been normalized, and its value range is [range missing]. .

[0047] As an example, the ratio of the slow blood flow index of any voxel to its local signal inhomogeneity, after normalization, is taken as the MRA signal reliability of that voxel. The normalization adopts the linear normalization method, and the normalization object is the ratio obtained from all voxels. At the same time, in order to avoid the denominator being 0 during the ratio calculation, a hyperparameter is added to the denominator. In this embodiment, the hyperparameter is described as 0.01.

[0048] Thus, the reliability of the MRA signal for each voxel was obtained.

[0049] Step S004: Based on the reliability of the CTA local signal and the reliability of the MRA signal, as well as the degree of modal contradiction of each voxel, assign weights to the two modal data and perform registration and fusion to realize multimodal three-dimensional reconstruction of blood vessels.

[0050] It should be noted that in order to effectively reconstruct the different vascular probabilities generated by the current voxel in the two modalities, the analysis weight of the voxel for different modalities is dynamically determined, that is, which modality is trusted more, and finally the unit reconstruction is performed reasonably.

[0051] Specifically, for any voxel, the modal data corresponding to the maximum value of the CTA local signal reliability and MRA signal reliability of that voxel is used as the reference modal data of that voxel; the sum of the CTA local signal reliability and MRA signal reliability of that voxel is obtained, and the ratio of the signal reliability corresponding to the reference modal data to the sum is used as the initial reference modal weight of that voxel; the sum obtained by adding 1 to the modal inconsistency degree of that voxel is multiplied by the initial reference modal weight, and the product is used as the reference modal weight of that voxel, and assigned to the corresponding modal data; at the same time, the difference obtained by subtracting the reference modal weight from 1 is assigned to another modal data other than the reference modal data. The specific modal weight calculation process is as follows: Figure 3 As shown, it uses CTA data as an example for calculation.

[0052] It should be noted that the signal with higher reliability among the two modal data is selected for the initial modal weight allocation, while also taking into account the impact of the degree of modal inconsistency.

[0053] Furthermore, for any voxel, based on the modal weights corresponding to the two modal data of that voxel, the representative vessel probabilities after the standardized processing of the two modal data of that voxel are weighted and fused to obtain the fused vessel probability of that voxel. This process is then used to obtain the fused vessel probabilities of each voxel, resulting in novel 3D volume data. The novel 3D volume data undergoes 3D smoothing processing using existing techniques, which will not be elaborated in this embodiment, to obtain a fused 3D volume. The fusion process is as follows: Figure 4 As shown.

[0054] It should be noted that the fused data integrates bimodal information, effectively mitigating the differences between the two in the original conflict area and providing an optimized data foundation for subsequent high-fidelity 3D reconstruction. At the same time, it avoids unnatural jumps in the reconstructed surface and achieves adaptive and interpretable intelligent allocation of fusion weights for each voxel, laying a solid foundation for accurate 3D reconstruction of cerebral blood vessels through multimodal fusion.

[0055] Furthermore, the fused 3D volume is essentially a fused weighted graph of two modal data. Applying it to the existing 3D reconstruction process using the Marching Cubes algorithm generates a vascular model represented by a triangular mesh, which is then provided to doctors as reliable auxiliary data for clinical diagnosis. The specific process involves using the weighted fused 3D volume as input and employing the Marching Cubes algorithm... The Cubes algorithm extracts initial isosurfaces to generate a vascular model represented by a triangular mesh. During this process, adaptive isosurface extraction is guided by a fused 3D volume: in regions with high weighted confidence and low inconsistency, a standard threshold is used to preserve anatomical details; in regions with low weighted confidence and high inconsistency, the isosurface threshold and smoothing coefficient are dynamically adjusted to suppress surface noise that may be introduced by modal conflicts. Furthermore, the reconstructed mesh is locally optimized based on the weight distribution: edge sharpening is enhanced in high CTA weighted regions to maintain the high spatial resolution of CTA, while curvature continuity constraints are applied to high MRA weighted regions to compensate for potential signal inhomogeneities. Simultaneously, combining vascular centerline topology and hemodynamic priors, local mesh subdivision and morphological modification are performed on stenotic regions to ensure that the lumen geometry conforms to physiological flow characteristics.

[0056] This concludes the embodiment.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A three-dimensional reconstruction method for cerebral vascular stenosis based on multimodal image fusion, characterized in that, The method includes the following steps: Acquire multimodal imaging data of the patient, including arterial phase CTA sequences and 3D TOF-MRA sequences, and obtain several voxels; Based on the difference in the probability of the same voxel representing blood vessels in different modal data, the modal contradiction degree of each voxel is obtained; several preliminary calcified voxels are screened based on the voxel CTA data, and the calcification influence factor of each voxel is obtained by combining the influence range of flowering artifact; considering the distribution relationship between the voxel CTA data and the vascular CT value performance, the CT value reliability of each voxel is obtained; and by combining the calcification influence factor and the CT value reliability, the CTA local signal reliability of each voxel is obtained. Based on the differences in MRA signal intensity within the vicinity of a voxel, the local signal inhomogeneity of each voxel is obtained; the vascular layer of each voxel is obtained through morphological analysis, and the signal intensity of different voxels at the same vascular layer is analyzed. Combined with the distribution of vascular layers, the slow blood flow index of each voxel is determined; the reliability of the MRA signal of each voxel is obtained by combining the local signal inhomogeneity and the slow blood flow index. Based on the reliability of the CTA local signal and the reliability of the MRA signal, as well as the degree of modal inconsistency of each voxel, weights are assigned to the two modal data and registration and fusion are performed.

2. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 1, characterized in that, The modal inconsistency degree of each voxel is obtained using the following method: The CTA and MRA data were standardized separately, and the converted data represented the probability of the representative vessels of each of the two modalities in each voxel. For any voxel, the absolute value of the difference between the two modal data and the representative blood vessel probability of that voxel is taken as the degree of modal contradiction of that voxel.

3. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 1, characterized in that, The specific methods for obtaining the calcification influencing factors of each voxel are as follows: For any voxel, obtain the Euclidean distance between the voxel and its nearest pre-calcified voxel. Use the ratio of the Euclidean distance to the defined radius of influence as the calcification distance factor of the voxel. Subtract the calcification distance factor of the voxel from 1 and output the result through a linear rectified function as the calcification influence factor of the voxel.

4. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 1, characterized in that, The specific methods for obtaining the reliability of the CT values ​​of each voxel are as follows: The CTA data of each voxel are used as the CT value of each voxel; preset calcification threshold, initial vascular threshold and ideal vascular threshold are set. If the CT value of any voxel is less than or equal to the initial vascular threshold, or greater than or equal to the calcification threshold, the confidence level of the CT value of that voxel is set to 0. If the CT value of the voxel is greater than the initial vascular threshold and less than or equal to the ideal vascular threshold, the difference between the CT value of the voxel and the initial vascular threshold is taken as the first difference, the difference between the ideal vascular threshold and the initial vascular threshold is taken as the first standard deviation, and the ratio of the first difference to the first standard deviation is taken as the confidence level of the CT value of the voxel. If the CT value of the voxel is greater than the ideal vascular threshold and less than the calcification threshold, the difference between the CT value of the voxel and the calcification threshold is taken as the second difference. The difference between the ideal vascular threshold and the calcification threshold is taken as the second standard deviation. The ratio of the second difference to the second standard deviation is taken as the confidence level of the CT value of the voxel.

5. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 1, characterized in that, The specific method for obtaining the CTA local signal reliability of each voxel is as follows: The difference obtained by subtracting the calcification influence factor of any voxel from 1, multiplied by the product of the CT value confidence of that voxel, is used as the CTA local signal reliability of that voxel.

6. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 1, characterized in that, The local signal non-uniformity of each voxel is obtained using the following method: The signal intensity corresponding to each voxel in the 3D TOF-MRA data is used as the MRA signal intensity of each voxel; the initial vascular mask is obtained; for any voxel, a spatial neighborhood of the voxel is constructed with the voxel as the center and the voxel in a range of 3×3×3, and the spatial neighborhood contains the voxel. The coefficient of variation of the MRA signal intensity of voxels belonging to the initial vascular mask in the spatial neighborhood of the voxel is used as the local signal inhomogeneity of the voxel.

7. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 1, characterized in that, The specific methods for obtaining the vascular layers of each voxel through morphological analysis include: By refining the morphology, a rough vascular centerline skeleton is extracted. For any voxel, based on the vascular centerline skeleton of the nearest voxel to it, the voxel is taken as the voxel of that segment of the blood vessel. The vascular tree structure is obtained by segmenting and tracing blood vessels. The branches of the vascular tree structure are numbered from small to large along the branch direction starting from the main trunk of the blood vessel, and each segment of the blood vessel is assigned a corresponding level. Thus, the level of the blood vessel corresponding to the voxel is obtained, which is used as the vascular level of the voxel.

8. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 1, characterized in that, The slow blood flow index of each voxel is obtained using the following method: For any voxel, obtain the mean MRA signal intensity of all voxels in the vascular layer where the voxel is located, and use the ratio of the voxel's MRA signal intensity to the mean MRA signal intensity as the signal deviation factor of the voxel; multiply the difference obtained by subtracting the signal deviation factor from 1 by the product of the ratio of the voxel's vascular layer to the maximum value among all vascular layers, and use the product as the slow flow index of the voxel.

9. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 2, characterized in that, The method for assigning weights and performing registration and fusion of the two modal data based on the reliability of the CTA local signal and the reliability of the MRA signal, as well as the degree of modal inconsistency of each voxel, includes the following specific methods: For any voxel, the modal data corresponding to the maximum value between the CTA local signal reliability and the MRA signal reliability of that voxel is used as the reference modal data of that voxel. Obtain the sum of the CTA local signal reliability and MRA signal reliability of the voxel, and use the ratio of the signal reliability corresponding to the reference modal data to the sum as the initial reference modal weight of the voxel. The sum of 1 and the modal contradiction degree of the voxel is multiplied by the product of the initial reference modal weights, and this product is used as the reference modal weight of the voxel and assigned to the corresponding modal data. Based on the two-modal data of each voxel and their corresponding weights, a fused 3D volume is obtained.

10. The method for three-dimensional reconstruction of cerebral vascular stenosis based on multimodal image fusion according to claim 9, characterized in that, The method for fusing the two-modal data of each voxel and their corresponding weights to obtain a fused 3D volume includes the following specific methods: For any voxel, based on the modal weights corresponding to the two modal data of that voxel, the representative vessel probabilities after the two modal data of that voxel are standardized are weighted and fused to obtain the fused vessel probability of that voxel. The fusion vessel probability of each voxel is obtained to get new three-dimensional volume data; the new three-dimensional volume data is then subjected to three-dimensional smoothing to obtain the fused three-dimensional volume.

Images