Automatic evaluation method for stenosis degree based on vascular image features

Abstract

This invention discloses an automatic assessment method for stenosis based on vascular imaging features, belonging to the field of medical image processing and automatic assessment of cardiovascular diseases. By receiving a dynamic vascular image sequence without respiratory motion correction and a synchronous respiratory signal, the method segments the target vascular segment in each frame and extracts parameters such as length, curvature, and diameter to generate a corresponding vascular deformation time-series curve. Then, phase correlation analysis is performed between the deformation curve and the respiratory motion signal to calculate the phase lock value, and the vascular compliance level is automatically determined by combining the diameter fluctuation amplitude. Finally, a vascular compliance index atlas is generated and output. This invention transforms respiratory motion from an interference signal into a natural load for vascular function assessment, achieving non-invasive, automatic, and quantitative assessment of vascular wall elasticity and blood flow reserve function. It overcomes the limitations of existing technologies that rely on morphological stenosis or complex invasive measurements, providing more comprehensive diagnostic evidence for vascular health in clinical practice.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and automatic assessment of cardiovascular diseases, and in particular to an automatic assessment method for the degree of stenosis based on vascular imaging features. Background Technology

[0002] Cardiovascular disease is the leading cause of death worldwide, and its core pathological basis lies in arterial stenosis and degeneration of vascular wall function. Currently, the clinical diagnosis of vascular stenosis mainly focuses on two aspects: morphological assessment and functional assessment.

[0003] Morphological assessment, centered on DSA and CTA, directly observes the degree of stenosis in the vascular lumen through imaging and is currently the basic diagnostic method. However, clinical practice shows that relying solely on the stenosis rate is insufficient to accurately reflect the impact of the lesion on blood flow reserve—lesions with the same degree of stenosis may exhibit different hemodynamic states due to differences in vessel wall elasticity, easily leading to missed diagnoses or overtreatment.

[0004] Functional assessment uses invasive fractional flow reserve (FFR) as the gold standard. Measuring the pressure ratio using a pressure guidewire to determine the degree of blood flow restriction is crucial for guiding revascularization. However, this technique is high-risk, expensive, and has contraindications, making it difficult to widely implement. While the non-invasive CT-FFR technique developed for this purpose can calculate FFR based on CTA images, it requires extremely high image quality, rigorously eliminating respiratory motion artifacts, and is computationally complex and time-consuming, hindering its widespread adoption at the grassroots level.

[0005] In the field of vascular image processing, vascular deformation caused by respiratory motion has long been considered an image artifact that must be eliminated. Driven by this concept, a series of technologies have been developed globally, including respiratory gating, motion correction algorithms, and AI artifact elimination, all with the core goal of acquiring static vascular images.

[0006] However, this "de-respiratory" approach overlooks a crucial functional indicator of blood vessels as living tissue—vascular compliance. This parameter directly reflects the elasticity of the blood vessel wall and its blood flow reserve capacity. Changes in intrathoracic pressure caused by respiration drive rhythmic deformation of blood vessels, a characteristic that directly reflects the elasticity of the blood vessel wall. Current technology treats respiratory motion entirely as interference, failing to recognize its potential as a natural, non-invasive signal for assessing vascular compliance. This confines the industry to a static morphological assessment framework, hindering its ability to overcome the inherent limitations of existing functional techniques. Summary of the Invention

[0007] This invention overcomes the shortcomings of the prior art and provides an automatic assessment method for the degree of stenosis based on vascular imaging features.

[0008] To achieve the above objectives, the technical solution adopted by this invention is: an automatic assessment method for the degree of stenosis based on vascular imaging features, comprising the following steps:

[0009] S1. Receive a dynamic vascular image sequence of multiple consecutive respiratory cycles, and a respiratory motion signal acquired synchronously with the dynamic vascular image sequence, wherein the dynamic vascular image sequence is an initial image sequence that has not undergone respiratory motion correction processing and retains the vascular morphological changes caused by respiratory motion.

[0010] S2. Perform target blood vessel segmentation processing on each frame of the dynamic blood vessel image sequence, extract the center line of the target blood vessel segment in each frame, and calculate the length parameter, curvature parameter, and diameter parameter of the target blood vessel segment corresponding to each frame based on the center line.

[0011] S3. Based on the temporal changes of the length, curvature, and diameter parameters of the target blood vessel segment with the respiratory cycle, generate a time-series curve of blood vessel deformation corresponding to the three parameters; perform phase correlation analysis on the time-series curve of blood vessel deformation and the respiratory motion signal, and calculate the phase lock value characterizing the degree of synchronization between blood vessel deformation and respiratory motion.

[0012] S4. Based on the phase lock value and the fluctuation range of the diameter parameter of the target blood vessel segment with the respiratory cycle, the vascular compliance level of the target blood vessel segment is determined.

[0013] S5. Based on the vascular compliance level determination result of the target vascular segment, generate and output a vascular compliance index map.

[0014] In a preferred embodiment of the present invention, a dynamic vascular image sequence covering at least three complete respiratory cycles is received from a medical imaging acquisition device; the dynamic vascular image sequence is a digital subtraction angiography (DSA) sequence or a CT angiography (CTA) sequence; simultaneously, a respiratory motion signal aligned with the timestamp of the dynamic vascular image sequence is received; the respiratory motion signal is a diaphragmatic movement timing signal or a thoracic expansion timing signal; the time synchronization deviation between each frame of the dynamic vascular image sequence and the respiratory motion signal is verified, and image frames with a time synchronization deviation less than a preset threshold and their corresponding respiratory motion signal data are retained.

[0015] In a preferred embodiment of the present invention, the calculation process for the length parameter, curvature parameter, and pipe diameter parameter includes:

[0016] S201. Using a deep learning model for blood vessel segmentation, each frame of the dynamic blood vessel image sequence is segmented into target blood vessel segments at the pixel level, and a binary mask of the target blood vessel segments is generated.

[0017] S202. Based on the binarized mask, extract the sub-pixel precision centerline of the target blood vessel segment, and calculate the diameter parameter of the target blood vessel segment in each frame image along the normal direction of the centerline; at the same time, calculate the length parameter and curvature parameter of the target blood vessel segment in the corresponding frame image based on the coordinate sequence of the centerline.

[0018] In a preferred embodiment of the present invention, the process of generating the vascular deformation time-series curve includes:

[0019] S301. Perform time-series alignment on the length, curvature, and diameter parameters of the target blood vessel segment corresponding to each frame of the dynamic vascular image sequence.

[0020] S302, Based on the normalized interval, perform analysis on the length parameter, curvature parameter, and pipe diameter parameter. Normalization processing;

[0021] S303. Remove outliers by sliding window filtering, and smooth the data by polynomial fitting to generate a vascular deformation time-series curve corresponding to the target vascular segment that has the same time-series length as the respiratory motion signal.

[0022] In a preferred embodiment of the present invention, the calculation process of the phase-locked value includes:

[0023] S311. Perform Hilbert transform on the vascular deformation time-series curve and the respiratory motion signal respectively to obtain the instantaneous phase sequences corresponding to the two sets of signals;

[0024] S312. Calculate the instantaneous phase difference between the two sets of instantaneous phase sequences, and calculate the standard deviation of the instantaneous phase difference over the entire respiratory cycle;

[0025] S313. Based on the standard deviation, a phase-lock value characterizing the synchronization between vascular deformation and respiratory movement is calculated, wherein the phase-lock value ranges from 0 to 1.

[0026] In a preferred embodiment of the present invention, the process of determining the vascular compliance level includes:

[0027] S401. Calculate the maximum and minimum values ​​of the diameter parameter of the target blood vessel segment during a complete respiratory cycle, and calculate the fluctuation range of the diameter parameter with the respiratory cycle based on the maximum and minimum values.

[0028] S402. When the phase lock value is greater than the first threshold and the fluctuation range of the tube diameter parameter is greater than 8%, the target blood vessel segment is determined to be of high compliance level; when the phase lock value is less than the second threshold or the fluctuation range of the tube diameter parameter is less than 2%, the target blood vessel segment is determined to be of low compliance level; otherwise, the target blood vessel segment is determined to be of medium compliance level.

[0029] In a preferred embodiment of the present invention, the process of generating the vascular compliance index map is as follows:

[0030] S501. Spatial registration is performed between the vascular compliance level, the corresponding phase lock value, and the fluctuation amplitude of the diameter parameter of the target vascular segment and the corresponding frame image of the dynamic vascular image sequence.

[0031] S502. Color mapping is performed on vascular segments with different vascular compliance levels using color coding rules. The color mapping results are superimposed on the vascular anatomy images of the dynamic vascular image sequence to generate and output a vascular compliance index map containing vascular anatomy and corresponding compliance indicators.

[0032] In a preferred embodiment of the present invention, when verifying the time synchronization deviation between each frame of the dynamic vascular imaging sequence and the respiratory motion signal, the sampling data with the closest timestamp in the respiratory motion signal is matched based on the acquisition timestamp of each frame of the dynamic vascular imaging sequence.

[0033] Calculate the difference between the acquisition timestamp of each image frame and the timestamp of the matching respiratory motion signal sampling data, and use the difference as the time synchronization deviation; retain image frames and corresponding respiratory motion signal sampling data with a time synchronization deviation of less than 50ms, and discard image frames and corresponding data that do not meet the requirements.

[0034] In a preferred embodiment of the present invention, after determining the vascular compliance level of the target vascular segment, the morphological stenosis parameter of the target vascular segment is extracted. The morphological stenosis parameter is the ratio of the minimum diameter of the target vascular segment to the normal reference diameter at the proximal end.

[0035] The vascular compliance level of the target vascular segment is correlated and matched with the morphological stenosis parameter to generate vascular blood flow reserve function assessment data corresponding to the target vascular segment, and the vascular blood flow reserve function assessment data is added to the vascular compliance index atlas.

[0036] In a preferred embodiment of the present invention, when the determination result is the high compliance level, a vascular compliance index map is generated, indicating good wall elasticity and sufficient blood flow reserve.

[0037] When the determination result is the low compliance level, a vascular compliance index map is generated, indicating vascular wall stiffness and impaired hemodynamic reserve.

[0038] This invention addresses the shortcomings of the prior art and has the following beneficial effects:

[0039] (1) By receiving dynamic vascular image sequences without respiratory motion correction and simultaneously acquiring respiratory motion signals, the dynamic deformation information of vascular length, curvature, and diameter during the respiratory cycle is fully preserved. Based on the phase correlation analysis between these deformation characteristics and respiratory motion, the phase lock value between vascular deformation and respiratory motion is quantitatively calculated. Combined with the diameter fluctuation amplitude, the vascular compliance level is automatically determined, and finally an intuitive vascular compliance index map is generated. This not only avoids the operational risks and high costs of traditional functional assessment methods, but also eliminates the need to rely on non-invasive blood flow simulation technology with extremely high image quality requirements and complex calculations. While retaining the routine clinical image acquisition process, it achieves non-invasive, automatic, and quantitative assessment of vascular wall elasticity and blood flow reserve function, significantly improving the accessibility and clinical applicability of functional diagnosis of vascular stenosis.

[0040] (2) By extracting multidimensional deformation parameters such as length, curvature, and diameter of the vascular segment during the complete respiratory cycle, and calculating its instantaneous phase lock value with respiratory motion based on Hilbert transform, the response characteristics of the vascular wall to changes in respiratory pressure can be accurately captured, thereby effectively distinguishing vascular segments with different elastic states. When the phase lock value is high and the diameter fluctuation is significant, it is judged as a high-compliance vessel, indicating good wall elasticity and sufficient blood flow reserve; conversely, when the phase lock value is low or the diameter fluctuation is weak, it is judged as a low-compliance vessel, indicating rigid wall and impaired hemodynamic reserve. This dual judgment mechanism based on deformation characteristics and respiratory motion synchronization overcomes the deficiency of simply relying on morphological stenosis rate to reflect the blood flow function status, and can identify vascular segments with insignificant morphological stenosis but impaired function (such as rigid functional stenosis), providing a more comprehensive basis for vascular health assessment in clinical practice.

[0041] (3) By spatially registering core parameters such as vascular compliance grade, phase lock value, and diameter fluctuation amplitude with vascular anatomy, and generating a vascular compliance index atlas using color coding rules, the functional assessment results are presented intuitively and visually on anatomical images. Clinicians can easily identify vascular segments with different compliance grades and their spatial distribution through the atlas, and simultaneously obtain morphological stenosis rate and blood flow reserve function assessment data for each vascular segment, significantly improving the interpretability of diagnostic information and the efficiency of clinical decision-making. This method is compatible with routinely acquired DSA and CTA image data, requires no additional equipment or complex operations, has good clinical adaptability and promotional value, and can effectively fill the gap in the field of vascular function assessment in existing technologies, promoting the leapfrog development of cardiovascular disease diagnosis from static morphology to dynamic function. Attached Figure Description

[0042] 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 recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of a preferred embodiment of the present invention; Detailed Implementation

[0044] 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.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein. Therefore, the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0046] The automatic assessment method for stenosis degree based on vascular imaging features provided in this embodiment is executed by a computer processor capable of parallel processing of medical image data. The overall process is implemented based on the time-sequential execution logic of a computer program, with machine-side data processing as the core, fully covering all technical features of the claims. The core of this invention addresses the common technical bias in existing vascular imaging diagnosis: that is, the need to eliminate respiratory motion artifacts for accurate diagnosis. It overcomes the limitation of relying solely on morphological stenosis degree to judge vascular lesions by converting respiratory motion from interference signals into diagnostic signals, and combines this with vascular compliance function indicators to achieve automatic functional assessment of vascular stenosis degree. Specific implementation details are as follows:

[0047] like Figure 1 As shown, the automatic assessment method for the degree of stenosis based on vascular imaging features includes the following steps:

[0048] S1. Receive dynamic vascular image sequences of multiple consecutive respiratory cycles, as well as respiratory motion signals acquired synchronously with the dynamic vascular image sequences. The dynamic vascular image sequences are initial image sequences that have not undergone respiratory motion correction processing and retain the vascular morphological changes caused by respiratory motion.

[0049] Furthermore, the system receives a dynamic vascular image sequence transmitted from a medical imaging acquisition device, covering no less than three complete respiratory cycles; the dynamic vascular image sequence is a digital subtraction angiography (DSA) sequence or a CT angiography (CTA) sequence; simultaneously, it receives respiratory motion signals aligned with the timestamps of the dynamic vascular image sequence; the respiratory motion signals are diaphragmatic movement timing signals or thoracic expansion timing signals; it verifies the time synchronization deviation between each frame of the dynamic vascular image sequence and the respiratory motion signals, and retains image frames with time synchronization deviations less than a preset threshold and their corresponding respiratory motion signal data.

[0050] When verifying the time synchronization deviation between each frame of the dynamic vascular imaging sequence and the respiratory motion signal, the acquisition timestamp of each frame of the dynamic vascular imaging sequence is used as the benchmark to match the sampling data with the closest timestamp in the respiratory motion signal.

[0051] Calculate the difference between the acquisition timestamp of each image frame and the timestamp of the matched respiratory motion signal sampling data, and use the difference as the time synchronization deviation; retain image frames and corresponding respiratory motion signal sampling data with a time synchronization deviation of less than 50ms, and discard image frames and corresponding data that do not meet the requirements.

[0052] In practical implementation, the first steps are to select the target vessel segment, receive initial data, and perform synchronous verification. For the target vessel segment, this embodiment provides two parallel selection rules: The first is a manual selection mode, which receives the vessel segment to be evaluated defined by the user through the interactive interface, including clinically relevant target vessel areas such as coronary arteries and peripheral arteries. The defined area must cover the complete start and end points of the vessel segment without branch interference. The second is an automatic identification mode, which uses a vascular topology recognition algorithm to automatically extract the complete topological network of the main vessel and its branches from the dynamic vascular image, selecting vessel segments with continuous lumens, no obvious branch crossings, and a length of not less than 20 mm as the target vessel segment.

[0053] After identifying the target vascular segment, the system receives a sequence of dynamic vascular images spanning multiple respiratory cycles transmitted from the medical imaging acquisition equipment, along with respiratory motion signals acquired synchronously with the dynamic vascular image sequence. During the acquisition of the dynamic vascular image sequence, the equipment's built-in respiratory gating, breath-hold acquisition, and real-time motion correction functions must be disabled. After acquisition, no post-processing is performed on the image sequence to eliminate respiratory motion artifacts or correct respiratory motion, thus fully preserving the vascular morphological changes caused by respiratory motion and fundamentally avoiding the loss of vascular compliance-related functional characteristics caused by de-respiratory processing in existing technologies. For the two imaging modalities, specific adaptation requirements are defined: the two-dimensional DSA sequence must be a single-plane projection image sequence with an acquisition frame rate of no less than 10fps, an acquisition duration covering no less than three complete respiratory cycles, and a single frame resolution of no less than 512×512 pixels. The three-dimensional CTA volumetric sequence is a dynamic volumetric image sequence with a temporal resolution of not less than 100ms / volumetric frame, an acquisition duration covering not less than 3 complete respiratory cycles, a spatial resolution of not less than 0.5mm×0.5mm×0.5mm for a single volumetric image, and a slice thickness of not more than 1mm.

[0054] Dynamic vascular imaging sequences can be obtained using digital subtraction angiography (DSA) or CT angiography (CTA) sequences. The acquisition duration of the image sequence must cover at least three complete respiratory cycles to ensure statistical significance in the correlation analysis between respiratory rhythm and vascular deformation. The acquisition time interval of a single frame should not exceed 100 ms to meet the temporal resolution requirements of time series analysis. Simultaneously, respiratory motion signals strictly aligned with the timestamps of the dynamic vascular imaging sequence should be received. These respiratory motion signals can be diaphragmatic motion time series signals or thoracic expansion time series signals, with a sampling frequency of at least 20 Hz to ensure accurate temporal matching with the dynamic vascular imaging sequence.

[0055] After data reception is complete, the respiratory cycle identification and segmentation operation is performed first to establish a unified time reference for subsequent time series analysis: based on the respiratory motion signal, after removing baseline drift and high-frequency noise using bandpass filtering, the inspiratory peak and expiratory trough in the signal are identified using a peak detection algorithm. The time interval between two adjacent inspiratory peaks is taken as a complete respiratory cycle, and those exceeding the time limit are discarded. Abnormal respiratory cycles within a preset range are selected, and at least three consecutive, valid, and complete respiratory cycles are retained as a unified time reference for all subsequent time-series analyses and parameter calculations. A time synchronization verification operation between the image frames and respiratory motion signals is then performed.

[0056] In practical implementation, the acquisition timestamp of the i-th frame in the dynamic vascular imaging sequence is used. Based on this, the sampling point with the closest timestamp is matched in the sampling data of respiratory motion signals. Calculate the time synchronization deviation between the two. ,in This refers to the sequence number of the image frame. N is the total number of frames in the dynamic vascular imaging sequence. This refers to the sequence number of the respiratory motion signal sampling point; the preset time synchronization deviation threshold is 50ms, and only [data] is retained. The image frames and corresponding respiratory motion signal sampling data are collected, and invalid data that does not meet the synchronization requirements are removed to ensure the timing accuracy of subsequent phase analysis.

[0057] After receiving the data, the respiratory cycle identification and segmentation operation is performed first to establish a unified time reference for subsequent time series analysis. The specific implementation is as follows:

[0058] A fourth-order Butterworth bandpass filter was used for filtering, with upper and lower cutoff frequencies set to 0.5Hz and 0.1Hz, respectively, to match the normal human respiratory rate range. Simultaneously, a second-order polynomial fitting was used to remove baseline drift, resulting in the preprocessed respiratory motion signal. ;

[0059] An adaptive peak detection algorithm was used to identify inspiratory peaks and expiratory troughs in the preprocessed respiratory motion signal. The peak amplitude threshold was set to 20% of the global peak value of the signal, and the minimum interval between adjacent inspiratory peaks was set to 1.5s. False peaks caused by noise were eliminated by the positive and negative changes in the signal gradient before and after the peak. The time interval between two adjacent inspiratory peaks was taken as a complete respiratory cycle to complete the division of the respiratory cycle of the entire sequence.

[0060] Excluding single-cycle durations exceeding the 2s-8s range and respiratory rate variation coefficients The rhythmic irregularities and shallow breathing cycles with peak values ​​less than 10% of the signal baseline were identified. Ultimately, at least three consecutive, valid, and complete breathing cycles were retained as a unified time reference for all subsequent time series analyses and parameter calculations.

[0061] After completing the respiratory cycle segmentation, respiratory-vascular deformation physiological delay correction is performed to eliminate the influence of the inherent physiological delay between thoracic / diaphragmatic movement and vascular wall deformation on phase analysis. Specifically, this is implemented by: processing the preprocessed respiratory motion signal... The pipe diameter deformation curve generated in subsequent steps Perform cross-correlation analysis; the cross-correlation function is calculated using the following formula: In the formula, The time shift parameter ranges from -200ms to 200ms (covering common clinical physiological delays), and N is the total number of sequence sampling points; the cross-correlation function is used. The maximum value corresponding to Value, as a physiological delay between respiratory movements and vascular deformation ; respiratory movement signals Translate along the time axis The corrected respiratory motion signal was obtained. This eliminates phase system errors caused by physiological delay.

[0062] S2. Perform target vessel segmentation processing on each frame of the dynamic vascular image sequence, extract the centerline of the target vessel segment in each frame, and calculate the length parameter, curvature parameter, and diameter parameter of the target vessel segment corresponding to each frame based on the centerline.

[0063] Furthermore, the calculation process for length parameters, curvature parameters, and pipe diameter parameters includes:

[0064] S201. Using a deep learning model for blood vessel segmentation, pixel-level segmentation of the target blood vessel segment is performed on each frame of the dynamic blood vessel image sequence to generate a binary mask of the target blood vessel segment.

[0065] S202. Extract the sub-pixel precision centerline of the target blood vessel segment based on the binarized mask, and calculate the diameter parameter of the target blood vessel segment in each frame of the image along the normal direction of the centerline; at the same time, calculate the length parameter and curvature parameter of the target blood vessel segment in the corresponding frame of the image based on the coordinate sequence of the centerline.

[0066] In practice, a pre-trained deep learning model for blood vessel segmentation is used to segment the target blood vessel segments at the pixel level for each frame of the image, and corresponding segmentation logic is adapted for both 2D DSA and 3D CTA sequences:

[0067] Two-dimensional DSA sequence segmentation: A pre-trained two-dimensional U-Net++ semantic segmentation model was used. The input was a single-frame grayscale image cropped from a pre-defined target vessel segment ROI. Image preprocessing included grayscale normalization. The dimensions are uniformly scaled to 512×512 pixels. The model output is a probability map of the target vessel segment. The probability map is binarized using a classification threshold of 0.5 to obtain an initial binary mask. Morphological opening operations (with a 3×3 circular kernel as the structuring element) are performed on the initial mask to remove isolated small connected components, and then morphological closing operations are performed to fill the holes in the lumen. Finally, through connected component analysis, only the largest connected component that matches the ROI space of the target vessel segment is retained to obtain the final binary mask of the target vessel segment. ,in These are the pixel coordinates of the image frame. This indicates that the pixel belongs to the target blood vessel segment region. This indicates that the pixel belongs to the background area.

[0068] 3D CTA sequence segmentation: A pre-trained 3D U-Net++ semantic segmentation model was used. The input was a 3D volumetric image cropped from a pre-defined target vessel segment ROI. Image preprocessing included adjusting the HU value window width and level (coronary artery sequences were set to -200HU to 800HU, peripheral artery sequences to 0HU to 1000HU) and uniformly scaling the image size to 256×256×128 voxels. The model output was a 3D probability map of the target vessel segment. The probability map was binarized using a classification threshold of 0.5 to obtain an initial 3D binarized mask. A 3D morphological opening operation (using a 3×3×3 spherical kernel as the structuring element) was performed on the initial mask to remove isolated small connected components, followed by a 3D morphological closing operation to fill the pores within the lumen. Finally, through 3D connected component analysis, only the largest connected component spatially matching the target vessel segment ROI was retained, resulting in the final 3D binarized mask of the target vessel segment. ,in For the voxel coordinates of the volumetric image, This indicates that the voxel belongs to the target vascular segment region. This indicates that the voxel belongs to the background region.

[0069] The deep learning model for vascular segmentation can use the U-Net series of semantic segmentation networks. The training dataset of the model contains more than 10,000 DSA and CTA image samples with annotated coronary artery and peripheral artery vascular regions. The samples cover clinical cases of different ages and different degrees of vascular lesions. During the training process, data augmentation methods such as random flipping, rotation, and grayscale scaling are used to improve the model's generalization ability.

[0070] Based on the generated binary mask Extract the centerline of the target blood vessel segment with sub-pixel / sub-voxel precision, and adapt the corresponding extraction logic for 2D / 3D images respectively:

[0071] Centerline extraction from 2D DSA sequences: The Zhang-Suen thinning algorithm is used to extract the skeleton of the 2D mask region of the target vessel segment, resulting in a pixel-level centerline skeleton. Post-processing is performed on the pixel-level skeleton: spurious branches with a length of less than 3 pixels are removed, irrelevant branches that do not conform to the direction of the target vessel segment are removed, and the start and end points of the centerline are trimmed to be completely aligned with the pre-defined start and end points of the target vessel segment. The post-processed pixel-level centerline coordinate sequence is fitted with a cubic B-spline curve. During the fitting process, the least squares method is used to optimize the fitting residual. The fitting residual threshold is set to 0.01 pixels, and control points are selected at 1 every 5 pixels, finally obtaining a sub-pixel precision centerline coordinate sequence. ,in This represents the k-th sub-pixel coordinate point of the center line, with a coordinate precision of 0.01 pixels. This represents the total number of coordinate points along the center line.

[0072] Centerline extraction of 3D CTA sequence: A distance-transform-based 3D thinning algorithm is used to extract the skeleton of the 3D mask region of the target vessel segment, obtaining a voxel-level centerline skeleton. Post-processing is performed on the voxel-level skeleton: spurious branches with a length <3 voxels are removed, irrelevant branches that do not conform to the direction of the target vessel segment are removed, and the start and end points of the centerline are trimmed to be completely aligned with the pre-defined start and end points of the target vessel segment. The post-processed voxel-level centerline coordinate sequence is fitted with a 3D cubic B-spline curve, with the fitting residual threshold set to 0.01 voxels. Control points are selected at 1 point every 5 voxels, finally obtaining a sub-voxel precision 3D centerline coordinate sequence. ,in The first of the center lines Sub-voxel coordinates, with a coordinate precision of 0.01 voxels. This represents the total number of coordinate points along the center line.

[0073] Automatic repair is implemented to address breakage issues that occur during centerline extraction: if the breakage length... 5mm, use linear interpolation to complete the centerline coordinates at the fracture point; if the fracture length is... If the value is 5mm, the centerline extraction is deemed a failure, and the user is prompted to manually correct the target vessel segment ROI and then re-execute the extraction process.

[0074] Based on the centerline coordinate sequence, the length, curvature, and diameter parameters of the target blood vessel segment in each frame of the image are calculated. The corresponding calculation formulas are adapted for two-dimensional and three-dimensional images, and the boundary processing rules for diameter calculation are clarified to ensure the accuracy of parameter calculation.

[0075] Length parameter calculation: The length parameter of the target vessel segment is the cumulative Euclidean distance of the centerline coordinate sequence, and the calculation formula is:

[0076] Two-dimensional sequence: ;

[0077] Three-dimensional sequence: ;

[0078] In the formula, , , For the center line The x, y, and depth coordinates of each coordinate point. , , For the center line The corresponding coordinate values ​​of each coordinate point.

[0079] Curvature parameter calculation: Curvature parameters of the target vessel segment Let be the average curvature of each coordinate point on the centerline, for the th ... coordinate points Its curvature The calculation formula is:

[0080] Two-dimensional sequence: ;

[0081] Three-dimensional sequence: ;

[0082] In the formula, in the two-dimensional sequence , for The first derivative of the point coordinates, , for Second derivative of point coordinates; in a three-dimensional sequence for The first derivative of the position vector of a point (tangent vector). The second derivative of the position vector. This is the vector cross product operator. This is a vector magnitude operator; all derivatives are obtained by differentiating the results after fitting a third-order polynomial to the centerline coordinate sequence. The final curvature parameter of the target vessel segment. .

[0083] Diameter parameter calculation (including boundary treatment rules): The diameter parameter D of the target vessel segment is calculated along the normal direction / normal plane of the centerline, while strictly applying boundary treatment rules, specifically as follows:

[0084] Two-dimensional sequence pipe diameter calculation: for the k-th coordinate point of the centerline Calculate the tangent vector of the centerline at this point, and solve for the normal direction perpendicular to the blood vessel's orientation based on the tangent vector. Traverse the boundaries of the blood vessel region in the binary mask along the normal direction to both sides, and use the Zernike moment sub-pixel edge detection algorithm to obtain the boundary points of the inner and outer walls of the blood vessel with sub-pixel precision. Calculate the Euclidean distance between the two boundary points, which is the blood vessel diameter at that point. .

[0085] Three-dimensional sequence pipe diameter calculation: for the k-th coordinate point of the centerline Calculate the tangent vector of the centerline at this point, and construct a normal plane perpendicular to the vessel's orientation based on the tangent vector. Within the normal plane, traverse the vessel region boundary of the 3D mask along 16 uniformly selected directions within 360°. Use a 3D sub-pixel edge detection algorithm to obtain sub-voxel precision boundary points, and calculate the distance from the boundary points in all directions to the vessel's orientation. The average value of twice the distance between points is the diameter of the blood vessel at that point. .

[0086] Interference exclusion rules: For non-luminal boundaries such as vascular branches, plaque protrusions, and calcifications encountered during the normal / normal plane traversal, the interference is excluded by judging the continuity of vascular topology: if the distance between a boundary point in a certain direction and the boundary point in an adjacent direction exceeds 20% of the average diameter, it is judged as an abnormal boundary point, the measurement value in that direction is removed, and the measurement value in the adjacent normal direction is used as the interpolation replacement.

[0087] Final pipe diameter calculation: After removing outlier measurements, the pipe diameter is calculated at all coordinate points along the centerline. The average value is used as the diameter parameter of the target blood vessel segment in that frame of image. .

[0088] S3. Based on the temporal changes of the length, curvature, and diameter parameters of the target vascular segment with the respiratory cycle, generate a vascular deformation time-series curve corresponding to the three parameters; perform phase correlation analysis between the vascular deformation time-series curve and the respiratory motion signal to calculate the phase lock value characterizing the synchronization between vascular deformation and respiratory motion.

[0089] Furthermore, the generation process of the vascular deformation time series curve includes:

[0090] S301. Perform time-series alignment of the length, curvature, and diameter parameters of the target blood vessel segment corresponding to each frame of the dynamic vascular image sequence.

[0091] S302, Based on the normalized interval, perform analysis on the length parameter, curvature parameter, and pipe diameter parameter. Normalization processing;

[0092] S303. Remove outliers by sliding window filtering and smooth the curve by polynomial fitting to generate a vascular deformation time-series curve corresponding to the target vascular segment with the same time-series length as the respiratory motion signal.

[0093] Furthermore, the calculation process for the phase-locked value includes:

[0094] S311. Perform Hilbert transform on the vascular deformation time-series curve and the respiratory motion signal respectively to obtain the instantaneous phase sequences corresponding to the two sets of signals;

[0095] S312. Calculate the instantaneous phase difference between the two sets of instantaneous phase sequences, and calculate the standard deviation of the instantaneous phase difference over the entire respiratory cycle.

[0096] S313. Based on the standard deviation, the phase lock value, which characterizes the degree of synchronization between vascular deformation and respiratory movement, is calculated. The phase lock value ranges from 0 to 1.

[0097] In practice, after calculating the length, curvature, and diameter parameters of the target blood vessel segment corresponding to each frame of image, a corresponding time-series curve of blood vessel deformation is generated based on the temporal changes of each parameter with the respiratory cycle. The specific implementation is as follows:

[0098] Temporal alignment: The length, curvature, and tube diameter parameters corresponding to each image frame are linearly interpolated according to the acquisition timestamp of the image frame and mapped to the corrected respiratory motion signal. On a unified time axis, the sampling frequency and duration of the parameter time sequence are made completely consistent with those of the respiratory motion signal to form the initial parameter time sequence.

[0099] Normalization: Time series with any parameter Where n is the total number of sampling points in the time series, and the minimum value of the sequence over all valid complete respiratory cycles is used. Maximum value As a normalization interval, Normalization process, normalized sequence The calculation formula is: In the formula, Let be the parameter value of the t-th sampling point in the time series.

[0100] Smoothing and Denoising: A median sliding window filter with a window size of 5 sampling points is used to remove outliers in the normalized parameter time series and eliminate abnormal data caused by image noise and segmentation errors. For the filtered sequence within each effective complete respiratory cycle, a global third-order polynomial fitting method is used for smoothing, with the fitting residual threshold set to 0.01. Finally, a vascular deformation time series curve that is completely consistent with the respiratory motion signal time series length is generated. The vascular deformation time series curve includes a length deformation curve, a curvature deformation curve, and a diameter deformation curve, which correspond to the time series changes of the length parameter, curvature parameter, and diameter parameter with the respiratory cycle, respectively. The subsequent phase correlation analysis takes the diameter deformation curve as the core analysis object.

[0101] After generating the vascular deformation time-series curve, phase correlation analysis is performed between the vascular deformation time-series curve and the corrected respiratory motion signal to calculate the phase lock value (PLV), which characterizes the degree of synchronization between vascular deformation and respiratory motion. The specific implementation is as follows:

[0102] Signal preprocessing: pipe diameter deformation curve With corrected respiratory motion signals Then, a second-order polynomial detrending and a 0.1~0.5Hz bandpass filter are performed again to eliminate the interference of signal non-stationarity and high-frequency noise on the Hilbert transform.

[0103] Instantaneous phase solution: for the pre-processed pipe diameter deformation curve Respiratory movement signals Perform Hilbert transforms on each signal to obtain the corresponding analytic signals: .

[0104] In the formula, For Hilbert transform operators, The imaginary unit is used; the instantaneous phase sequence of two sets of signals is calculated based on the analytic signal, and the instantaneous phase of the pipe diameter deformation curve is obtained. Instantaneous phase of respiratory motion signal ,in An operator for solving complex arguments.

[0105] Phase unwrapping: Calculating the instantaneous phase difference between two sets of instantaneous phase sequences. A gradient-based phase unwrapping algorithm is used to... The sequence is unwrapped to eliminate the limitations caused by the phase value range. The resulting phase jump yields a continuous sequence of instantaneous phase differences after unwrapping. .

[0106] Phase lock value calculation: For each valid complete respiratory cycle, the sequence within the cycle is resampled to a uniform set of 100 sampling points. Based on the resampled unwrapped phase difference sequence, the phase lock value for a single cycle is calculated. The calculation formula is: .

[0107] In the formula, It is a natural constant. To solve for the complex modulus operator; take the single cycle of all valid complete respiratory cycles. The arithmetic mean of the values ​​is used as the final phase-locked value (PLV). The PLV ranges from 0 to 1. The closer the PLV is to 1, the higher the degree of synchronization between vascular deformation and respiratory movement, and the better the phase-locked effect. The closer the PLV is to 0, the lower the degree of synchronization between the two, and the worse the phase-locked effect.

[0108] S4. Based on the phase lock value and the fluctuation range of the target vascular segment's diameter parameters with the respiratory cycle, the vascular compliance level of the target vascular segment is determined.

[0109] Furthermore, the process of determining vascular compliance levels includes:

[0110] S401. Calculate the maximum and minimum values ​​of the diameter parameters of the target blood vessel segment during the complete respiratory cycle, and calculate the fluctuation range of the diameter parameters with the respiratory cycle based on the maximum and minimum values.

[0111] S402. When the phase lock value is greater than the first threshold and the fluctuation range of the tube diameter parameter is greater than 8%, the target vessel segment is determined to be of high compliance level; when the phase lock value is less than the second threshold or the fluctuation range of the tube diameter parameter is less than 2%, the target vessel segment is determined to be of low compliance level; otherwise, the target vessel segment is determined to be of medium compliance level.

[0112] After determining the vascular compliance level of the target vessel segment, the morphological stenosis parameter of the target vessel segment is extracted. The morphological stenosis parameter is the ratio of the minimum diameter of the target vessel segment to the normal reference diameter at the proximal end.

[0113] The vascular compliance level of the target vessel segment is correlated and matched with the morphological stenosis parameter to generate vascular flow reserve function assessment data corresponding to the target vessel segment, and the vascular flow reserve function assessment data is added to the vascular compliance index atlas.

[0114] In practice, after obtaining the phase-locked value (PLV), the fluctuation range of the target vessel diameter parameter with the respiratory cycle is calculated. Specifically, the maximum value of the diameter parameter within each effective complete respiratory cycle is extracted. and minimum value Calculate the diameter fluctuation amplitude within a single respiratory cycle, and take the average of the diameter fluctuation amplitudes within all valid complete respiratory cycles as the final diameter fluctuation amplitude. The calculation formula is: .

[0115] In the formula, The degree of change in tubing diameter with respiratory motion is represented as a percentage. This is based on the calculated phase-locked value (PLV) and the tubing diameter fluctuation amplitude. The vascular compliance level of the target vessel segment is determined, with a preset first phase-locking threshold of 0.85 and a second phase-locking threshold of 0.3. The rules for handling boundary conditions and mutual exclusion conditions are also defined, as follows:

[0116] Core judgment rule: When PLV 0.85 and A_D At 8%, the target vascular segment is classified as having a high compliance level.

[0117] When PLV 0.3 or At 2%, the target vascular segment is classified as having a low compliance level.

[0118] In other cases, the target vessel segment is classified as having medium compliance.

[0119] Threshold boundary handling rule: When PLV=0.85 and When the compliance rate is 8%, it is classified as a high compliance level.

[0120] When PLV=0.3 or When the compliance rate is 2%, it is classified as a low compliance level.

[0121] Mutual exclusion condition secondary judgment rule: for PLV>0.85 but <2%, PLV<0.3 but For mutual exclusion exceptions exceeding 8%, a second judgment is performed:

[0122] Calculate the phase lock-in values ​​corresponding to the length deformation curve and the curvature deformation curve respectively. If two or more of the phase lock-in values ​​corresponding to the three parameters meet the criteria for a high compliance level, the system is classified as high compliance. If two or more meet the criteria for a low compliance level, the system is classified as low compliance. All other cases are classified as medium compliance.

[0123] After determining the vascular compliance level, and combining vascular morphology parameters, an automatic assessment of the degree of vascular stenosis corresponding to the invention's subject matter is achieved. First, the technical definitions of core terms are clarified:

[0124] Blood flow restrictive stenosis: refers to stenosis in which vascular lesions reduce the blood flow reserve capacity of the target organ, making it unable to meet the blood supply demand under load. Borderline blood flow restrictive stenosis: refers to stenosis in which there is no obvious blood supply deficiency at rest, but blood supply to the target organ may be restricted under load.

[0125] The specific assessment implementation steps are as follows:

[0126] Morphological stenosis rate calculation: Extract the morphological stenosis degree parameter of the target vessel segment, i.e., the diameter stenosis rate. The calculation formula is: ,in The minimum diameter of the target vessel segment to be evaluated. The average diameter of a normal blood vessel segment that is at least 10 mm proximal to the target vessel segment and has no plaque or luminal narrowing, is defined as the segment whose proximal end is at least 10 mm from the stenotic area to be evaluated.

[0127] Special Case Reference Diameter Treatment: For diffuse vascular lesions (where no suitable proximal normal reference segment exists across the entire vessel segment), the following priority rules are used to determine the reference diameter. :

[0128] The average diameter of a normal blood vessel segment with no lesions at the distal end of the target vessel segment.

[0129] Reference values ​​for normal vessel diameter in the same age group with the same vessel segment.

[0130] The average initial diameter of blood vessels with no lesions at the opening.

[0131] Functional assessment rules for stenosis degree: Based on vascular compliance grade and morphological stenosis rate RS, establish clear corresponding assessment rules:

[0132] When the target vessel segment has a high compliance level: if RS < 70%, it is judged as non-flow-restricting stenosis, indicating that the stenosis has not caused damage to the blood flow reserve function.

[0133] If RS ≥ 70%, it is considered a borderline blood flow restrictive stenosis, and further clinical evaluation is required.

[0134] When the target vessel segment is of medium compliance grade: if RS < 50%, it is judged as non-flow restrictive stenosis.

[0135] If 50% ≤ RS < 70%, it is considered a critical blood flow restrictive stenosis.

[0136] If RS ≥ 70%, it is considered as blood flow restrictive stenosis, indicating that vascular stenosis has caused damage to blood flow reserve function.

[0137] When the target vessel segment has a low compliance grade, it is considered a flow-restricting stenosis regardless of whether the morphological stenosis rate RS reaches 50%.

[0138] Low-compliance vessel segments without morphological stenosis (RS=0), after excluding temporary factors such as vasospasm and external compression, were labeled as rigid-walled functional stenosis, indicating functional blood flow restriction caused by vascular wall fibrosis and decreased elasticity. The automatic assessment results of the above stenosis degree were correlated and stored with the vessel compliance grade and core calculation parameters for subsequent atlas generation.

[0139] By correlating and matching the vascular compliance level of the target vessel segment with morphological stenosis parameters, assessment data of vascular flow reserve function corresponding to the target vessel segment is generated. Specifically, when the target vessel segment has a high compliance level, assessment data of good vessel wall elasticity and sufficient flow reserve function is generated; when the target vessel segment has a low compliance level, assessment data of stiff vessel wall and impaired hemodynamic reserve is generated. This achieves stenosis degree assessment based on vascular function, overcoming the technical limitations of existing technologies that rely solely on morphological stenosis judgment.

[0140] S5. Based on the vascular compliance level determination results of the target vascular segment, generate and output the vascular compliance index map.

[0141] Furthermore, the generation process of the vascular compliance index map is as follows:

[0142] S501. Spatial registration is performed between the vascular compliance level, corresponding phase lock value, and diameter parameter fluctuation amplitude of the target vascular segment and the corresponding frame image of the dynamic vascular imaging sequence.

[0143] S502. Color mapping is performed on vascular segments with different vascular compliance levels using color coding rules. The color mapping results are superimposed on the vascular anatomy images of the dynamic vascular imaging sequence to generate and output a vascular compliance index map containing vascular anatomy and corresponding compliance indicators.

[0144] In practice, the vascular compliance level of the target vessel segment, the corresponding phase lock value (PLV), and the diameter fluctuation range will be used. Morphological stenosis parameters and blood flow reserve function assessment data are spatially registered with corresponding frames of dynamic vascular imaging sequences. Using the centerline coordinate sequence of the target vascular segment as the registration benchmark, all assessment parameters are mapped to the spatial location of the corresponding vascular segment to ensure accurate spatial correspondence between parameters and vascular anatomy.

[0145] A preset color coding rule is used to color-map vessel segments with different compliance levels: high compliance segments are mapped to green, medium compliance segments to yellow, and low compliance segments to red. Simultaneously, the phase-locked value (PLV) and the diameter fluctuation amplitude are used as the color coding rules. The value is used to adjust the color transparency of the corresponding blood vessel segment. The closer the parameter value is to the judgment threshold, the higher the color transparency, so as to intuitively present the spatial distribution differences of blood vessel compliance.

[0146] The color mapping results are superimposed on the vascular anatomy images of the dynamic vascular imaging sequence. At the same time, the phase lock value, diameter fluctuation amplitude, stenosis rate and blood flow reserve function assessment results of each vascular segment are marked at the corresponding positions in the image. Finally, a vascular compliance index map containing vascular anatomy and corresponding compliance function indicators is generated and output, realizing the transformation of respiratory motion information, which is traditionally regarded as interference, into the core diagnostic basis for vascular health assessment.

[0147] After completing the determination of vascular compliance level and automatic assessment of stenosis degree, a vascular compliance index map is generated and output. First, the vascular compliance level of the target vessel segment, the corresponding phase-locked value (PLV), and the diameter fluctuation amplitude are calculated. The morphological stenosis rate (RS) and functional assessment results of stenosis degree were spatially registered with the corresponding frames of the dynamic vascular imaging sequence. Using the sub-pixel precision centerline coordinates of the target vessel segment as the registration benchmark, all assessment parameters were mapped to the spatial location of the vessel segment corresponding to the centerline, ensuring a complete match between the spatial location of each assessment parameter and the corresponding vascular anatomy. Subsequently, a preset green-yellow-red three-color coding rule was used to color-map vessel segments with different vascular compliance levels: high compliance levels were mapped to green (RGB values: 0, 255, 0), medium compliance levels to yellow (RGB values: 255, 255, 0), and low compliance levels to red (RGB values: 255, 0, 0). Simultaneously, the color transparency of the corresponding vessel segment was adjusted based on the core assessment parameters. Make adjustments. The value range is 0.2 to 1.0, and the specific calculation formula is as follows:

[0148] High compliance level: ,when hour, It decreases linearly with decreasing PLV, reaching a minimum of 0.2; when hour, ;

[0149] Low compliance level: ,when hour, It decreases linearly with increasing PLV, reaching a minimum of 0.2; when hour, ;

[0150] Medium compliance level: .

[0151] The color mapping results are superimposed onto the vascular anatomy images of the dynamic vascular imaging sequence. At the same time, the phase lock value, diameter fluctuation amplitude, stenosis rate and stenosis degree functional assessment results are marked at the corresponding vascular segment positions in the images. Finally, a vascular compliance index atlas is generated and output, which includes vascular anatomy, compliance spatial distribution, core assessment parameters and stenosis degree functional diagnosis results. This realizes the transformation of respiratory motion information, which is traditionally regarded as interference, into the core diagnostic basis for vascular health assessment.

[0152] To address frequently occurring anomalies during clinical data collection, a comprehensive anomaly identification and fallback mechanism has been established, as detailed below:

[0153] Image quality anomaly handling: For dynamic vascular imaging sequences, peak signal-to-noise ratio is used. The image was determined to be of low quality. Non-local mean denoising was first performed. Even after denoising, the PSNR remained low. The system prompts the user to re-acquire images; for cases of uneven contrast agent filling, it uses the gray-scale variance of the vascular region. If a vessel segment is determined to be unevenly filled, it will be automatically removed, and only the segments with uniform filling will be retained for subsequent analysis.

[0154] Abnormal processing of respiratory motion signals: targeting the signal-to-noise ratio of respiratory signals Effective and complete respiratory cycle In case of signal quality abnormalities, the system automatically performs alternative signal extraction: based on the movement trajectory of the diaphragm in the dynamic image sequence, it extracts the temporal displacement signal of the diaphragm as an alternative respiratory motion signal, and re-executes the respiratory cycle division, physiological delay correction and synchronization verification process; if the alternative signal still cannot meet the effective cycle requirements, it prompts the user to re-acquire the synchronized respiratory signal.

[0155] Anomaly handling in segmentation and centerline extraction: Addressing the length of connected vessel components in the segmentation mask. If 80% of the target vessel segment's defined length is reached, the segmentation is considered a failure. The ROI range is automatically expanded, and the segmentation is re-executed. If it still fails, the user is prompted to manually correct the ROI. (Regarding the centerline breakage length...) If the error is mm and cannot be automatically corrected, it is determined that the centerline extraction has failed, and the user is prompted to manually correct the centerline before continuing the subsequent process.

[0156] Based on the preferred embodiments of the present invention described above, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for automatic assessment of stenosis degree based on features of blood vessel images, characterized by, Includes the following steps: S1. Receive a dynamic vascular image sequence of multiple consecutive respiratory cycles, and a respiratory motion signal acquired synchronously with the dynamic vascular image sequence, wherein the dynamic vascular image sequence is an initial image sequence that has not undergone respiratory motion correction processing and retains the vascular morphological changes caused by respiratory motion. S2. Perform target blood vessel segmentation processing on each frame of the dynamic blood vessel image sequence, extract the center line of the target blood vessel segment in each frame, and calculate the length parameter, curvature parameter, and diameter parameter of the target blood vessel segment corresponding to each frame based on the center line. S3. Based on the temporal changes of the length, curvature, and diameter parameters of the target blood vessel segment with the respiratory cycle, generate time-series curves of blood vessel deformation corresponding to the three parameters respectively; perform phase correlation analysis on the blood vessel deformation time-series curves and the respiratory motion signal to calculate the phase lock value characterizing the degree of synchronization between blood vessel deformation and respiratory motion. S4. Based on the phase lock value and the fluctuation range of the diameter parameter of the target blood vessel segment with the respiratory cycle, the vascular compliance level of the target blood vessel segment is determined. S5. Based on the vascular compliance level determination result of the target vascular segment, generate and output a vascular compliance index map.

2. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 1, characterized in that: The system receives a dynamic vascular image sequence transmitted from a medical imaging acquisition device, covering no less than three complete respiratory cycles; the dynamic vascular image sequence is a digital subtraction angiography (DSA) sequence or a CT angiography (CTA) sequence; simultaneously, it receives respiratory motion signals aligned with the timestamps of the dynamic vascular image sequence; the respiratory motion signals are diaphragmatic movement timing signals or thoracic expansion timing signals; it verifies the time synchronization deviation between each frame of the dynamic vascular image sequence and the respiratory motion signals, and retains image frames with time synchronization deviations less than a preset threshold and their corresponding respiratory motion signal data.

3. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 1, characterized in that: The calculation process for the length parameter, curvature parameter, and pipe diameter parameter includes: S201. Using a deep learning model for blood vessel segmentation, each frame of the dynamic blood vessel image sequence is segmented into target blood vessel segments at the pixel level, and a binary mask of the target blood vessel segments is generated. S202. Based on the binarized mask, extract the sub-pixel precision centerline of the target blood vessel segment, and calculate the diameter parameter of the target blood vessel segment in each frame image along the normal direction of the centerline; at the same time, calculate the length parameter and curvature parameter of the target blood vessel segment in the corresponding frame image based on the coordinate sequence of the centerline.

4. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 1, characterized in that: The process of generating the time-series curve of vascular deformation includes: S301. Perform time-series alignment on the length, curvature, and diameter parameters of the target blood vessel segment corresponding to each frame of the dynamic vascular image sequence. S302, Based on the normalized interval, perform analysis on the length parameter, curvature parameter, and pipe diameter parameter. Normalization processing; S303. Remove outliers by sliding window filtering, and smooth the data by polynomial fitting to generate a vascular deformation time-series curve corresponding to the target vascular segment that has the same time-series length as the respiratory motion signal.

5. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 1, characterized in that: The calculation process for the phase lock value includes: S311. Perform Hilbert transform on the vascular deformation time-series curve and the respiratory motion signal respectively to obtain the instantaneous phase sequences corresponding to the two sets of signals; S312. Calculate the instantaneous phase difference between the two sets of instantaneous phase sequences, and calculate the standard deviation of the instantaneous phase difference over the entire respiratory cycle; S313. Based on the standard deviation, a phase-lock value characterizing the synchronization between vascular deformation and respiratory movement is calculated, wherein the phase-lock value ranges from 0 to 1.

6. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 1, characterized in that: The process for determining the vascular compliance level includes: S401. Calculate the maximum and minimum values ​​of the diameter parameter of the target blood vessel segment during a complete respiratory cycle, and calculate the fluctuation range of the diameter parameter with the respiratory cycle based on the maximum and minimum values. S402. When the phase lock value is greater than the first threshold and the fluctuation range of the tube diameter parameter is greater than 8%, the target blood vessel segment is determined to be of high compliance level; when the phase lock value is less than the second threshold or the fluctuation range of the tube diameter parameter is less than 2%, the target blood vessel segment is determined to be of low compliance level; otherwise, the target blood vessel segment is determined to be of medium compliance level.

7. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 1, characterized in that: The process of generating the vascular compliance index map is as follows: S501. Spatial registration is performed between the vascular compliance level, the corresponding phase lock value, and the fluctuation amplitude of the diameter parameter of the target vascular segment and the corresponding frame image of the dynamic vascular image sequence. S502. Color mapping is performed on vascular segments with different vascular compliance levels using color coding rules. The color mapping results are superimposed on the vascular anatomy images of the dynamic vascular image sequence to generate and output a vascular compliance index map containing vascular anatomy and corresponding compliance indicators.

8. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 2, characterized in that: When verifying the time synchronization deviation between each frame of the dynamic vascular imaging sequence and the respiratory motion signal, the sampling data with the closest timestamp in the respiratory motion signal is matched with the acquisition timestamp of each frame of the dynamic vascular imaging sequence as the reference. Calculate the difference between the acquisition timestamp of each image frame and the timestamp of the matching respiratory motion signal sampling data, and use the difference as the time synchronization deviation; retain image frames and corresponding respiratory motion signal sampling data with a time synchronization deviation of less than 50ms, and discard image frames and corresponding data that do not meet the requirements.

9. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 6, characterized in that: After determining the vascular compliance level of the target vascular segment, the morphological stenosis parameter of the target vascular segment is extracted. The morphological stenosis parameter is the ratio of the minimum diameter of the target vascular segment to the normal reference diameter at the proximal end. The vascular compliance level of the target vascular segment is correlated and matched with the morphological stenosis parameter to generate vascular blood flow reserve function assessment data corresponding to the target vascular segment, and the vascular blood flow reserve function assessment data is added to the vascular compliance index atlas.

10. The automatic assessment method for stenosis degree based on vascular imaging features according to claim 6, characterized in that: When the determination result is the high compliance level, a vascular compliance index map is generated, indicating good wall elasticity and sufficient blood flow reserve. When the determination result is the low compliance level, a vascular compliance index map is generated, indicating vascular wall stiffness and impaired hemodynamic reserve.

Images