An AI-based method and system for predicting coronary artery risk in the elderly
By using artificial intelligence methods to screen out the effects of coronary artery calcification and imaging abnormalities in the elderly population, reconstructing coronary artery lesion information, and matching changes in anatomical function, the problem of lesion quantification distortion in the prediction of coronary artery risk in the elderly is solved, and more accurate coronary artery risk assessment is achieved.
Patent Information
- Application Number
- CN202611618239.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
In the current technology for predicting coronary artery risk in the elderly, calcification affects the identification of lumen boundaries, fluctuations in cardiac motion and image quality lead to distorted quantification of coronary artery lesions, and the degree of coronary artery anatomical stenosis does not correlate with the effect on blood flow function, making it difficult to accurately distinguish different coronary artery lesion states.
Using an artificial intelligence-based approach, this study extracts coronary artery structural data, identifies vascular segment readability, verifies interfered vascular segments, and performs corresponding analysis of coronary artery anatomy and function. This process filters out unquantifiable coronary artery data, reconstructs reliable coronary artery lesion information, and distinguishes different coronary artery lesion patterns through spatial matching of coronary artery anatomy and functional changes.
It improves the reliability of coronary lesion data, avoids unreliable lesion quantification results, accurately distinguishes between focal and diffuse functional impairment lesions, and enhances the accuracy of coronary risk prediction in the elderly.
Smart Images

Figure CN123118636A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image data processing technology, specifically to an artificial intelligence-based method and system for predicting coronary artery risk in the elderly. Background Technology
[0002] Coronary artery disease is one of the most common cardiovascular diseases among the elderly. Coronary CT angiography can non-invasively visualize the lumen, plaques, and stenosis of the coronary arteries and has been widely used for coronary lesion screening and risk assessment. With the development of artificial intelligence image recognition technology, existing technologies can automatically extract the coronary artery centerline, segment the lumen region, and identify plaques to achieve automated quantitative analysis of coronary lesions and further combine clinical data for coronary risk prediction.
[0003] However, coronary arteries in elderly subjects often exhibit significant calcification. The high-density images generated by calcified areas can easily affect the identification of the true lumen boundaries. Furthermore, cardiac motion, insufficient local angiographic filling, and image quality fluctuations can also distort the lumen structure. Current automated coronary artery analysis methods typically calculate the degree of stenosis directly after lumen segmentation. When local vessel segments lack quantitative data on coronary lesions, they may still output definitive lesion results, leading to the inclusion of image-interferenced abnormal lumen information in subsequent risk prediction processes.
[0004] Furthermore, the degree of anatomical stenosis in the coronary arteries does not always correlate with the actual impact on blood flow function. Some localized stenosis may not result in obvious functional abnormalities, while multiple consecutive mild to moderate lesions may lead to persistent impairment of blood flow function. Risk assessment based solely on the maximum degree of stenosis, plaque burden, or a single functional parameter makes it difficult to accurately distinguish different coronary artery lesion states, thus affecting the reliability of coronary artery risk prediction results in the elderly. Therefore, identifying and addressing vessel segments that cannot be quantified for coronary artery lesions before coronary risk prediction, and establishing a reliable spatial correlation between coronary anatomical lesions and corresponding functional changes, are technical problems that need to be solved in current intelligent analysis of coronary arteries in the elderly. Summary of the Invention
[0005] To address the above situation and overcome the shortcomings of existing technologies, the present invention adopts the following technical solution: The present invention provides an artificial intelligence-based method for predicting coronary artery risk in the elderly, which includes the following steps:
[0006] Step S1: Extraction of coronary artery structure data;
[0007] Step S2: Coronary artery segment readability identification;
[0008] Step S3: Review of the affected vessel segment;
[0009] Step S4: Coronary artery anatomy and function correspondence analysis;
[0010] Step S5: Predicting coronary artery risk in the elderly.
[0011] Further, in step S1, the coronary artery structure data extraction is used to establish the data basis for coronary artery lesion analysis. Specifically, it involves obtaining coronary artery CT angiography data of elderly subjects, extracting the coronary artery centerline, lumen boundary and plaque region through an artificial intelligence image recognition model, dividing the coronary artery into continuous vessel segments according to the coronary artery centerline, establishing the spatial correspondence between each continuous vessel segment and the coronary artery CT angiography data, and obtaining coronary artery structure data.
[0012] Further, in step S2, the coronary artery segment readability identification is used to determine whether each continuous vessel segment has reliable imaging conditions for quantifying coronary artery lesions. Specifically, based on the coronary artery structure data, calcification boundary influence identification, continuous lumen structure verification, and local imaging anomaly identification are performed on each continuous vessel segment. Based on the identification results, the lesion analysis path of the corresponding continuous vessel segment is determined to obtain the coronary artery segment readability result data.
[0013] The interpretability identification of the coronary artery segment includes the following steps:
[0014] Step S21: Calcification boundary influence identification, used to determine the influence of calcification area on lumen boundary identification. Specifically, it involves extracting cross-sections corresponding to continuous vessel segments along the coronary artery centerline, identifying the location where calcification area covers or interferes with lumen boundary based on the positional relationship between calcification area and lumen boundary, and obtaining calcification boundary influence data.
[0015] Step S22: Continuous lumen structure verification, used to identify abnormal lumen changes in continuous vascular segments. Specifically, based on the calcification boundary influence data, the lumen position, lumen area and boundary morphology of adjacent cross sections are continuously compared to identify local abrupt changes that are inconsistent with the adjacent lumen structure, and abnormal lumen structure data are obtained.
[0016] Step S23: Local imaging anomaly identification, used to identify lumen interpretation anomalies caused by non-calcification factors. Specifically, based on the coronary artery structure data and lumen structure anomaly data, the continuity of the centerline of the continuous vessel segment, the lumen edge state and the angiographic filling state are verified to determine the location of the vessel with motion interference or angiographic filling abnormality, and local imaging anomaly data are obtained.
[0017] Step S24: Determine the vessel segment interpretation path, which is used to control the subsequent lesion analysis method of different continuous vessel segments. Specifically, based on the calcification boundary influence data, lumen structure abnormality data, and local imaging abnormality data, continuous vessel segments that meet the conditions for quantifying coronary artery lesions are determined as analyzable vessel segments. Continuous vessel segments with lumen structure abnormalities or local missing lumen boundaries affected by image interference and with corresponding alternative reconstruction data are determined as vessel segments to be reviewed. Continuous vessel segments that do not meet the conditions for quantifying coronary artery lesions are determined as unquantifiable vessel segments, thus obtaining the interpretable results data of coronary vessel segments.
[0018] Further, in step S3, the interference segment review is used to re-determine reliable coronary lesion information for the segment to be reviewed. Specifically, it involves determining the image interference type corresponding to the segment to be reviewed based on the interpretable coronary segment data, selecting an alternative reconstruction sequence from the reconstruction data corresponding to the same coronary CT angiography examination based on the image interference type, re-identifying the lumen boundary of the segment to be reviewed in the alternative reconstruction sequence, and performing consistency verification on the lumen identification results in different available reconstruction sequences. The segment to be reviewed that meets the consistency requirements is identified as a reviewable analyzable segment, and the segment to be reviewed that does not meet the consistency requirements is identified as an unquantifiable segment, thus obtaining the reviewed coronary lesion data.
[0019] Further, in step S4, the coronary anatomy and function correspondence analysis is used to determine the spatial correspondence between coronary anatomy lesions and changes in blood flow function. Specifically, it involves acquiring coronary function analysis data corresponding to the coronary CT angiography data, spatially registering the location of coronary anatomy lesions with the location of functional changes distributed along the coronary artery centerline based on the verified coronary lesion data and coronary function analysis data, and determining the coronary lesion pattern based on the positional correspondence and distribution status between the anatomy lesions and functional changes to obtain the coronary anatomy and function correspondence result data.
[0020] The coronary artery anatomy and function correspondence analysis includes the following steps:
[0021] Step S41: Coronary artery anatomical lesion localization, used to determine the reliable spatial range of coronary artery anatomical lesions, specifically, based on the verified coronary artery lesion data, mapping the analyzable vessel segment and the verified analyzable vessel segment to the corresponding coronary artery centerline, determining the start and end positions of each coronary artery lesion, and obtaining coronary artery anatomical lesion location data;
[0022] Step S42: Localization of coronary artery function changes, used to determine the spatial range of coronary artery function abnormalities. Specifically, the coronary artery function analysis data is mapped to the corresponding coronary artery centerline. Based on the continuous changes of functional parameters along the coronary artery centerline, the local functional change area and the continuous functional change area are determined to obtain the location data of coronary artery function changes.
[0023] Step S43: Anatomical function spatial matching, used to determine the correspondence between coronary artery anatomical lesions and coronary artery function changes. Specifically, based on the location data of the coronary artery anatomical lesions and the location data of the coronary artery function changes, spatial overlap analysis is performed on the anatomical lesion area and the functional change area. Anatomical lesions that meet the preset spatial correspondence conditions are associated with functional changes to obtain coronary artery anatomical function matching data.
[0024] Step S44: Coronary artery lesion pattern determination, used to distinguish the coronary artery lesion status corresponding to different anatomical-functional relationships. Specifically, based on the coronary artery anatomical-functional matching data, the vessel segment corresponding to the coronary artery lesion and local functional changes is identified as a focal function-related lesion, the vessel segment where the degree of anatomical lesion is inconsistent with the degree of corresponding functional changes is identified as an anatomical-functional inconsistent lesion, and the vessel corresponding to multiple consecutive coronary artery lesions and continuous functional changes is identified as a diffuse functional impairment lesion, thus obtaining the coronary artery anatomical-functional corresponding result data.
[0025] Further, in step S5, the elderly coronary artery risk prediction is used to determine the coronary artery risk of elderly test subjects based on reliable coronary artery lesions and their functional impact. Specifically, it involves obtaining basic clinical data of elderly test subjects, constructing coronary artery risk prediction features based on the basic clinical data, the reviewed coronary artery lesion data, and the corresponding results of coronary artery anatomy and function, and inputting them into a pre-trained artificial intelligence risk prediction model to perform coronary artery risk prediction, thereby obtaining elderly coronary artery risk prediction result data.
[0026] The present invention provides an artificial intelligence-based coronary artery risk prediction system for the elderly, including a coronary artery structure extraction module, a vessel segment interpretation module, a vessel segment verification module, an anatomical function correspondence module, and a coronary artery risk prediction module;
[0027] The coronary artery structure extraction module is used to extract coronary artery structure data, obtain coronary artery structure data, and send the coronary artery structure data to the vessel segment interpretation module.
[0028] The readable segment module is used for coronary artery segment readableness identification, obtaining coronary artery segment readableness result data, and sending the coronary artery segment readableness result data to the vascular segment verification module;
[0029] The vessel segment verification module is used to verify the interfered vessel segment, obtain the verified coronary artery lesion data, and send the verified coronary artery lesion data to the anatomical function corresponding module and the coronary artery risk prediction module.
[0030] The anatomical function correspondence module is used for coronary artery anatomical function correspondence analysis, to obtain coronary artery anatomical function correspondence result data, and to send the coronary artery anatomical function correspondence result data to the coronary artery risk prediction module;
[0031] The coronary artery risk prediction module is used for predicting coronary artery risk in the elderly, obtaining coronary artery risk prediction result data for the elderly, and outputting the coronary artery risk prediction result data for the elderly.
[0032] The beneficial effects achieved by adopting the above solution are as follows:
[0033] (1) To address the problem that coronary artery calcification and local imaging abnormalities in elderly subjects can easily lead to distortion in the quantification of coronary artery lesions, this scheme adopts a combination of identifiable vessel segment identification, verification of interfered vessel segments, and anatomical function correspondence analysis. By screening out coronary artery data that cannot be quantified before coronary artery risk prediction, and re-determining the coronary artery lesion information of vessel segments with verification conditions, the reliability of coronary artery lesion data used for artificial intelligence risk prediction is improved. For example, for vessel segments that are severely affected by calcification and whose lumen identification results are inconsistent between different reconstruction sequences, the stenosis results obtained from a single reconstruction sequence are not directly used for coronary artery risk prediction.
[0034] (2) To address the problem that calcification, motion interference, or abnormal angiography filling can lead to the inability to reliably determine the luminal boundary of some coronary artery segments, this solution adopts a segment-readable gating mechanism. By verifying the influence of calcification boundary, continuous lumen structure, and local imaging status, different segments are identified as analyzable segments, segments to be reviewed, and unquantifiable segments, respectively, thereby achieving pre-blocking of unreliable quantification results of coronary artery lesions. For example, when a local lumen mutation is accompanied by calcification boundary interference, the local lumen mutation is not directly identified as a true coronary artery stenosis.
[0035] (3) To address the inconsistency between the degree of coronary anatomical lesions and the actual impact on blood flow function, this scheme adopts an anatomical-functional spatial matching technology based on the coronary artery centerline. By spatially matching reliable coronary anatomical lesions with corresponding functional changes, and distinguishing different coronary lesion patterns according to the local or continuous distribution of functional changes, the scheme achieves the differentiation and identification of focal functional-related lesions, anatomical-functional inconsistent lesions, and diffuse functional impairment lesions. For example, when multiple consecutive coronary lesions correspond to continuous functional changes along the vessel, they are identified as diffuse functional impairment lesions, rather than relying solely on the maximum local stenosis for risk assessment. Attached Figure Description
[0036] Figure 1 A flowchart illustrating an artificial intelligence-based method for predicting coronary artery risk in the elderly, provided by this invention;
[0037] Figure 2 A schematic diagram of an artificial intelligence-based coronary artery risk prediction system for the elderly provided by the present invention;
[0038] Figure 3 This is a flowchart illustrating the process of identifying the interpretability of coronary artery segments in step S2.
[0039] Figure 4 This is a flowchart illustrating the coronary anatomy and function correspondence analysis in step S4.
[0040] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0041] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0042] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0043] Example 1, see Figure 1 This invention provides an artificial intelligence-based method for predicting coronary artery risk in the elderly, which includes the following steps:
[0044] Step S1: Extraction of coronary artery structure data;
[0045] Step S2: Coronary artery segment readability identification;
[0046] Step S3: Review of the affected vessel segment;
[0047] Step S4: Coronary artery anatomy and function correspondence analysis;
[0048] Step S5: Predicting coronary artery risk in the elderly.
[0049] By performing the above operations, this solution addresses the issue of distorted coronary lesion quantification caused by coronary artery calcification and local imaging abnormalities in elderly subjects. It employs a combination of techniques including identifiable vessel segment recognition, verification of interfered vessel segments, and anatomical function correspondence analysis. By filtering out coronary artery data that cannot be quantified before coronary risk prediction and re-determining coronary lesion information for vessel segments with verification conditions, the reliability of coronary lesion data used for artificial intelligence risk prediction is improved. For example, for vessel segments severely affected by calcification and where lumen identification results differ between different reconstruction sequences, the stenosis results obtained from a single reconstruction sequence are not directly used for coronary risk prediction.
[0050] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the coronary artery structure data extraction is used to establish the data basis for coronary artery lesion analysis. Specifically, it involves obtaining coronary artery CT angiography data of elderly subjects, extracting the coronary artery centerline, lumen boundary and plaque region through an artificial intelligence image recognition model, dividing the coronary artery into continuous vessel segments according to the coronary artery centerline, establishing the spatial correspondence between each continuous vessel segment and the coronary artery CT angiography data, and obtaining coronary artery structure data.
[0051] In specific implementation, in step S1, coronary artery CT angiography data of elderly subjects is acquired, and the corresponding pixel spacing, slice thickness and image spatial position parameters are read. The spatial scale of the coronary artery CT angiography data is unified according to the pixel spacing and slice thickness, so that the coronary artery structure in different layers can be located in a unified spatial coordinate system.
[0052] Preferably, the artificial intelligence image recognition model uses a three-dimensional image segmentation model trained with annotated samples from coronary CT angiography. The coronary artery region is determined by the artificial intelligence image recognition model, and the coronary artery centerline is extracted from the coronary artery region. A corresponding cross-section is generated using the coronary artery centerline as the positioning reference. The lumen boundary and plaque region are identified in each cross-section. For recognition results with missing boundaries or local positional jumps, the original state is retained, and no compensation is made using adjacent lumen data in step S1.
[0053] In practice, the coronary arteries are divided into continuous segments based on the branching positions and spatial continuity of the coronary artery centerline. The starting and ending positions of the centerline and the coverage area of each continuous segment are recorded. The spatial correspondence between each continuous segment and the coronary CT angiography data is established through these position records. This allows subsequent steps to retrieve the corresponding coronary CT angiography data based on the position of the continuous segments. The coronary artery centerline, lumen boundary, plaque region, and the spatial correspondence are collectively used as the coronary vascular structure data.
[0054] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the coronary artery segment readability identification is used to determine whether each continuous vessel segment has reliable imaging conditions for quantifying coronary artery lesions. Specifically, based on the coronary artery structure data, calcification boundary influence identification, continuous lumen structure verification, and local imaging anomaly identification are performed on each continuous vessel segment. Based on the identification results, the lesion analysis path of the corresponding continuous vessel segment is determined to obtain the coronary artery segment readability result data.
[0055] The interpretability identification of the coronary artery segment includes the following steps:
[0056] Step S21: Calcification boundary influence identification, used to determine the influence of calcification area on lumen boundary identification. Specifically, it involves extracting cross-sections corresponding to continuous vessel segments along the coronary artery centerline, identifying the location where calcification area covers or interferes with lumen boundary based on the positional relationship between calcification area and lumen boundary, and obtaining calcification boundary influence data.
[0057] In practice, cross-sections perpendicular to the local direction of the coronary artery centerline are generated along the coronary artery centerline at a uniform spatial sampling interval. The spatial sampling interval is determined based on the image voxel size after spatial scale unification in step S1. For example, when the image voxel size after spatial scale unification is 0.5 mm, cross-sections are extracted at intervals of 0.5 mm, and the coronary artery CT angiography data corresponding to each cross-section is read according to the spatial correspondence in the coronary artery structure data.
[0058] Preferably, the calcified region is determined based on the plaque region obtained in step S1; for the current cross-section, the nearest plaque-free cross-section is searched proximally along the corresponding coronary artery centerline, and the luminal boundary of the plaque-free cross-section is shrunk inward by a spatially unified image voxel size, and the CT values of each voxel in the shrunk luminal region are read, using the CT values as a local angiography reference; when there is no plaque-free cross-section within the corresponding coronary artery centerline branch, the nearest plaque-free cross-section is searched proximally along the next higher level coronary artery centerline of that branch; when there are no usable plaque-free cross-sections within the coronary artery centerline range, the CT value of the aortic root lumen region in the same coronary CT angiography data is used as a local angiography reference;
[0059] In practice, the mean and standard deviation of the CT value corresponding to the local angiography reference are calculated. The sum of the mean and twice the standard deviation is used as the calcification identification boundary. Voxels in the plaque area whose CT values reach the calcification identification boundary are subjected to three-dimensional connectivity analysis according to 26 neighborhoods. The resulting connected voxel regions are identified as calcified regions. The calcification identification boundary is used to distinguish high CT value regions of plaques that exceed the fluctuation range of local angiography enhancement. The local angiography reference reduces the impact of differences in angiography enhancement levels between different coronary CT angiography examinations on the identification of calcified regions.
[0060] In practice, the calcified region and the corresponding cross-sectional cavity boundary are mapped to the same cross-sectional coordinate system. For each boundary position on the cavity boundary, when the corresponding boundary position spatially overlaps with the calcified region, or when the shortest distance from the corresponding boundary position to the calcified region is not greater than the image voxel size after spatial scaling, the boundary position is determined as the calcification influence boundary position. Continuous calcification influence boundary positions are connected along the cavity boundary, and the corresponding boundary arc lengths are accumulated to obtain the calcification influence boundary length. The calcification boundary influence ratio of the k-th cross-section is calculated according to the following formula:
[0061] ;
[0062] in, This represents the proportion of the calcification boundary influence at the k-th cross section, and is a dimensionless quantity. This represents the length of the calcification influence boundary at the k-th cross-section, in millimeters. The superscript cal indicates the calcification influence. The total length of the lumen boundary at the k-th cross-section is expressed in millimeters; the subscript k indicates the cross-section number arranged sequentially along the coronary artery centerline of the corresponding continuous vascular segment; the calcification boundary influence ratio is used to characterize the proportion of the lumen boundary affected by the calcification region in the current cross-section to the total length of the lumen boundary.
[0063] As a further optimization of this embodiment, when the influence ratio of the calcification boundary reaches 0.15, the corresponding cross-section is determined as the location of the calcification boundary influence. The 0.15 is used to exclude isolated influences caused by a small number of calcified areas being locally close to the lumen boundary. When historical labeled data is available, the influence ratio of the calcification boundary corresponding to the cross-sections that can be continuously identified by the lumen boundary after manual verification can be statistically analyzed, and the 95th percentile of the distribution of the calcification boundary influence ratio is used as the judgment limit. The location of the calcification boundary influence, the corresponding calcification boundary influence ratio, and the spatial range of the calcification influence boundary are recorded together to obtain the calcification boundary influence data.
[0064] Step S22: Continuous lumen structure verification, used to identify abnormal lumen changes in continuous vascular segments. Specifically, based on the calcification boundary influence data, the lumen position, lumen area and boundary morphology of adjacent cross sections are continuously compared to identify local abrupt changes that are inconsistent with the adjacent lumen structure, and abnormal lumen structure data are obtained.
[0065] In practice, the location of the calcification boundary influence is determined based on the calcification boundary influence data, and the lumen boundary of each cross section is read sequentially along the coronary artery centerline direction to calculate the corresponding lumen center position and lumen area; at the same time, the lumen boundaries of adjacent cross sections are resampled at a uniform angular interval to compare the boundary morphology of adjacent cross sections.
[0066] Preferably, for the k-th cross-section, the degree of local area abrupt change is calculated based on the lumen area of its preceding and following adjacent cross-sections:
[0067] ;
[0068] in, It represents the degree of abrupt change in the local area of the k-th cross section, and is a dimensionless quantity; This represents the area of the tube cavity at the k-th cross-section, in square millimeters. This represents the area of the tube cavity in the previous cross-section, expressed in square millimeters. This indicates the area of the lumen at the next cross-section, in square millimeters; subscript k, and These represent the numbers of the current cross section, the previous cross section, and the next cross section, respectively; the degree of local area abrupt change is used to characterize the degree of deviation of the lumen area of the current cross section from the average lumen area of the adjacent cross sections;
[0069] In practice, when the abrupt change in local area reaches 0.25, an abnormal lumen area is identified; when the offset of the lumen center position of adjacent cross-sections in the direction perpendicular to the coronary artery centerline exceeds two unified voxel sizes, an abnormal lumen position is identified; for boundary morphology, the lumen boundaries of adjacent cross-sections are resampled at 5° intervals, and the corresponding boundary points are compared after aligning the lumen center positions. When the radial distance difference of more than 20% of the corresponding boundary points exceeds one unified voxel size, an abnormal boundary morphology is identified; when historical calibrated vessel segment data is available, the above judgment values can be corrected by using the 95th percentile of the normal distribution of the corresponding measurement.
[0070] It should be noted that historically calibrated vascular segment data refers to continuous vascular segment data in historical coronary CT angiography data that has been manually verified to ensure that the lumen boundary can be continuously identified and that there is no calcification boundary influence, motion interference, or angiographic filling abnormality; historically calibrated samples refer to historical elderly test subjects samples that contain the aforementioned historically calibrated vascular segment data and have corresponding coronary function analysis data or follow-up results.
[0071] As a further optimization of this embodiment, when the same cross section satisfies any two of the following: abnormal lumen area, abnormal lumen position, and abnormal boundary morphology, it is identified as a local mutation location; for locations where only the lumen area changes continuously without being accompanied by abnormal lumen position or abnormal boundary morphology, it is not directly identified as a local mutation location; the local mutation location, the corresponding anomaly type, and the degree of local area mutation are recorded as lumen structure anomaly data.
[0072] Step S23: Local imaging anomaly identification, used to identify lumen interpretation anomalies caused by non-calcification factors. Specifically, based on the coronary artery structure data and lumen structure anomaly data, the continuity of the centerline of the continuous vessel segment, the lumen edge state and the angiographic filling state are verified to determine the location of the vessel with motion interference or angiographic filling abnormality, and local imaging anomaly data are obtained.
[0073] In practice, the location of local mutation is determined based on the abnormal data of the lumen structure, and the CT angiography data of the corresponding location and its adjacent locations are called according to the spatial correspondence in the coronary artery structure data; for the remaining areas, continuous verification is performed along the coronary artery centerline in the same way.
[0074] Preferably, for the continuous state of the centerline, five consecutive coronary artery centerline sampling points are used as the verification interval. The local centerline change direction is fitted based on the centerline sampling points before and after the current sampling point. When the vertical deviation distance of the current sampling point relative to the local centerline change direction exceeds two voxel sizes after spatial scale unification, the corresponding position is determined as the abnormal position of the continuous centerline. For the lumen edge state, CT values within a voxel size range are extracted along the lumen boundary normal direction to the inner and outer sides of the lumen. When the difference in CT values on both sides of the boundary is less than 50% of the median difference in CT values corresponding to the stable cross sections of adjacent structures, and the length of the affected boundary exceeds 30% of the total length of the current lumen boundary, the corresponding position is determined as the abnormal position of the lumen edge.
[0075] The structurally stable cross section refers to a cross section that is not identified as a location affected by calcification boundaries and is not identified as a location of local abrupt change.
[0076] In practice, for the angiography filling state, the median CT value of the lumen inside the lumen boundary of each cross section is extracted, and the median CT value of the lumen of the stable cross sections of the adjacent structures at the proximal and distal ends of the current verification position is used as the local angiography reference; when the median CT value of the lumen of three consecutive cross sections is lower than 70% of the local angiography reference, and the corresponding position is not occupied by the calcified area, the corresponding blood vessel position is determined as the angiography filling abnormal position;
[0077] As a further optimization of this embodiment, when the interval between the continuous abnormal position of the centerline and the abnormal position of the lumen edge along the direction of the coronary artery centerline does not exceed one cross-sectional sampling interval, the corresponding vessel position is determined as the motion interference position; for positions where both motion interference and angiographic filling abnormality exist, the corresponding image interference type is recorded respectively; the motion interference position, the angiographic filling abnormality position and the corresponding centerline interval are recorded as local imaging abnormality data.
[0078] Step S24: Determine the interpretation path of the vessel segment. This is used to control the subsequent lesion analysis method of different continuous vessel segments. Specifically, based on the calcification boundary influence data, lumen structure abnormality data, and local imaging abnormality data, continuous vessel segments that meet the conditions for quantifying coronary artery lesions are determined as analyzable vessel segments. Continuous vessel segments with lumen structure abnormalities or local missing lumen boundaries affected by image interference and with corresponding alternative reconstruction data are determined as vessel segments to be reviewed. Continuous vessel segments that do not meet the conditions for quantifying coronary artery lesions are determined as unquantifiable vessel segments, thus obtaining the interpretable result data of coronary vessel segments.
[0079] In specific implementation, the calcification boundary influence location in the calcification boundary influence data, as well as the motion interference location and angiography filling abnormality location in the local imaging abnormality data, are mapped to the coronary artery centerline of the corresponding continuous vessel segment to form the image interference location; the local mutation location in the lumen structure abnormality data is mapped to the same coronary artery centerline. When the local mutation location and the image interference location spatially overlap, or when the distance between the two along the coronary artery centerline does not exceed one cross-sectional sampling interval, the corresponding location is determined as the lumen structure abnormality location affected by image interference.
[0080] Preferably, when there are no image interference locations in a continuous vascular segment, or when there are image interference locations but no abnormal luminal structure affected by image interference and the corresponding luminal boundary remains intact, the continuous vascular segment is determined as an analyzable vascular segment; for locations with only local mutation locations but no corresponding calcification boundary influence or local imaging abnormalities, the interpretability of the corresponding continuous vascular segment is not directly reduced due to the local mutation location.
[0081] In practice, when there are abnormal locations of the lumen structure in a continuous vascular segment affected by image interference, or when image interference causes local loss of the lumen boundary at the corresponding location, the catalog of reconstruction sequences corresponding to the same coronary CT angiography examination is retrieved; for motion interference, other cardiac cycle reconstruction sequences are checked; for calcification boundary effects, reconstruction sequences with different spatial resolution characteristics are checked; for angiographic filling abnormalities, other reconstruction data with effective lumen filling information at the corresponding vascular location are checked.
[0082] As a further optimization of this embodiment, when there are abnormal locations of the lumen structure affected by image interference or local missing parts of the lumen boundary, and there is alternative reconstruction data for the corresponding image interference type, the corresponding continuous vascular segment is identified as the vascular segment to be reviewed; when the above-mentioned image interference exists and there is no corresponding alternative reconstruction data, the corresponding continuous vascular segment is identified as the unquantifiable vascular segment; the status of the analyzable vascular segment, the vascular segment to be reviewed, or the unquantifiable vascular segment corresponding to each continuous vascular segment, as well as the corresponding abnormal location and image interference type, are recorded together to obtain the interpretable result data of the coronary vascular segment.
[0083] Example 4, see Figure 1 , Figure 2This embodiment is based on the above embodiment. In step S3, the interference vessel segment review is used to re-determine reliable coronary lesion information for the vessel segment to be reviewed. Specifically, it determines the image interference type corresponding to the vessel segment to be reviewed based on the interpretable result data of the coronary vessel segment, selects an alternative reconstruction sequence from the reconstruction data corresponding to the same coronary CT angiography examination based on the image interference type, re-identifies the lumen boundary of the vessel segment to be reviewed in the alternative reconstruction sequence, and performs consistency verification on the lumen identification results in different available reconstruction sequences. The vessel segment to be reviewed that meets the consistency requirements is determined as a reviewable analyzable vessel segment, and the vessel segment to be reviewed that does not meet the consistency requirements is determined as an unquantifiable vessel segment, thus obtaining the reviewed coronary lesion data.
[0084] In specific implementation, in step S3, the segment to be reviewed and its corresponding image interference type are determined based on the interpretable data of the coronary artery segment. When the segment to be reviewed has motion interference, the same coronary artery centerline interval is called from other cardiac cycle reconstruction sequences corresponding to the same coronary CT angiography examination, and is verified according to the same method of determining the continuity of the centerline and the state of the lumen edge in step S23. When the number of continuous abnormal positions of the centerline in the candidate reconstruction sequence is less than that in the original reconstruction sequence, and the proportion of the length of the affected boundary corresponding to the abnormal position of the lumen edge to the total length of the lumen boundary is lower than that in the original reconstruction sequence, the candidate reconstruction sequence is retained as the alternative reconstruction sequence for the corresponding motion interference.
[0085] Preferably, when the vascular segment to be reviewed is affected by calcification boundaries, the same coronary artery centerline interval is retrieved from the reconstruction data with different spatial resolution characteristics in the same examination, and the proportion of calcification boundary influence corresponding to each calcification boundary influence position in the original reconstruction sequence is calculated in the same way as in step S21; for candidate reconstruction sequences, the proportion of calcification boundary influence is recalculated at the same centerline position; when the average value of the proportion of calcification boundary influence corresponding to the candidate reconstruction sequence is lower than the average value corresponding to the original reconstruction sequence, the candidate reconstruction sequence is retained as the alternative reconstruction sequence corresponding to the calcification boundary influence.
[0086] In practice, when angiographic filling abnormalities are found in the segment of blood vessel to be reviewed, the same coronary artery centerline interval is retrieved from other reconstruction data corresponding to the same coronary CT angiography examination, and the same angiographic filling status determination method as in step S23 is used for verification; when the corresponding blood vessel position in the candidate reconstruction sequence no longer meets the angiographic filling abnormality determination conditions, the candidate reconstruction sequence is retained as the alternative reconstruction sequence for the corresponding angiographic filling abnormality; when two or more types of image interference exist simultaneously in the same segment of blood vessel to be reviewed, only the candidate reconstruction sequence that simultaneously meets the improvement conditions for each of the corresponding image interference types is retained as the alternative reconstruction sequence.
[0087] As a further optimization of this implementation, alternative reconstruction sequences that meet the corresponding image interference improvement conditions and can continuously obtain the lumen boundary within the coronary artery centerline interval corresponding to the segment to be reviewed are identified as usable reconstruction sequences; when the number of usable reconstruction sequences is less than 2, coronary lesion information is not directly determined using a single alternative reconstruction sequence, and the corresponding segment to be reviewed is identified as an unquantifiable segment; when the number of usable reconstruction sequences is not less than 2, the consistency of lumen identification results in different usable reconstruction sequences is verified.
[0088] Preferably, based on the spatial correspondence in the coronary artery structure data, the vascular segments to be verified in each available reconstruction sequence are located to the same coronary artery centerline interval, and the same artificial intelligence image recognition model as in step S1 is used to re-identify the lumen boundary and plaque region; after the re-identification is completed, the lumen center position and lumen area corresponding to the same centerline position in each available reconstruction sequence are compared pairwise, wherein the allowable deviation of the lumen center position is determined according to the original pixel spacing of the coronary artery CT angiography data, and in this embodiment, one original pixel spacing is used; the allowable difference of the lumen area is determined according to the relative difference of repeated measurements of vascular segments with multiple reconstruction sequences in different reconstruction sequences in the historical calibrated vascular segment data, and in this embodiment, the 95th percentile of the distribution of the relative difference of repeated measurements is used as the judgment limit for the allowable difference of the lumen area;
[0089] As a further optimization of this implementation method, the consistency ratio of the reconstruction results of the vessel segment to be reviewed is calculated based on the cross-sections involved in the consistency verification:
[0090] ;
[0091] in, This represents the proportion of consistent reconstruction results for the vascular segment to be reviewed, and is a dimensionless quantity. The term "consistency ratio" indicates the number of cross sections where, after pairwise comparison of all available reconstructed sequences in the corresponding cross sections, the maximum pairwise deviation of the lumen center position does not exceed the allowable deviation of the lumen center position, and the maximum pairwise relative difference of the lumen area does not exceed the allowable difference of the lumen area. The unit is "number". "N" represents the total number of cross sections where all available reconstructed sequences have completed lumen boundary identification and participated in consistency verification. The subscript "c" indicates the cross section that meets the consistency requirements. The consistency ratio of the reconstruction results is used to characterize the spatial proportion of the vascular segment to be reviewed that can stably obtain lumen identification results in different available reconstructed sequences.
[0092] In practical implementation, when the reconstruction results are consistent with the proportion of When the value reaches 0.90, the lumen identification results in different available reconstruction sequences are determined to meet the consistency requirements. The 0.90 is used to allow a small number of cross sections to be affected by local noise or reconstruction differences. When historically calibrated vessel segment data is available, the 5th percentile of the distribution of the consistency ratio of reconstruction results of historically calibrated vessel segments can be used as the judgment threshold. The vessel segments to be reviewed that meet the consistency requirements are determined as reviewable and analyzable vessel segments. When the consistency ratio of reconstruction results does not reach the corresponding judgment threshold, the vessel segments to be reviewed are determined as unquantifiable vessel segments, and the lumen identification results in any single reconstruction sequence are not used to replace the coronary lesion information of the vessel segment.
[0093] In specific implementation, for the analyzable vessel segment determined in step S2, the coronary lesion area is determined based on the plaque area obtained in step S1, and the lumen area is determined based on the corresponding lumen boundary; for the re-analyzable vessel segment, the coronary lesion area is determined based on the plaque area re-identified in different available reconstruction sequences, and the lumen area is determined based on the lumen boundary that meets the consistency requirements; for each coronary lesion area, the plaque-free and lumen-boundary intact cross section with the closest distance between its proximal and distal ends is used as the reference cross section. When there are reference cross sections on both sides, the average lumen area of the two reference cross sections is used as the reference lumen area. When there is only one reference cross section, the lumen area of the reference cross section on that side is used as the reference lumen area, and the degree of lumen stenosis is determined by the decrease ratio of the minimum lumen area in the coronary lesion area to the reference lumen area; when there are no available reference cross sections on both sides of the coronary lesion area, the determined degree of lumen stenosis is not output for the coronary lesion area, and the corresponding continuous vessel segment is adjusted to an unquantifiable vessel segment;
[0094] In practice, the spatial location and degree of stenosis of the coronary artery lesion corresponding to the analyzable vessel segment determined in step S2, the spatial location and degree of stenosis of the coronary artery lesion re-determined by the re-analyzable vessel segment, and the spatial location of the non-quantifiable vessel segment finally determined in steps S2 and S3 are jointly organized and recorded. The status of the analyzable vessel segment, the re-analyzable vessel segment, or the non-quantifiable vessel segment corresponding to each continuous vessel segment is recorded to obtain the re-verified coronary artery lesion data.
[0095] By performing the above operations, this solution addresses the problem of unreliable determination of the lumen boundary in some coronary artery segments due to calcification, motion interference, or abnormal angiographic filling. It employs a segment-readable gating mechanism. By verifying the influence of calcification boundaries, continuous lumen structure, and local imaging status, different segments are identified as analyzable segments, segments awaiting review, and unquantifiable segments, respectively, thus achieving pre-blocking of unreliable coronary lesion quantification results. For example, when a local lumen mutation is accompanied by calcification boundary interference, the local lumen mutation is not directly identified as true coronary artery stenosis.
[0096] Example 5, see Figure 1 , Figure 4 This embodiment is based on the above embodiment. In step S4, the coronary anatomy and function correspondence analysis is used to determine the spatial correspondence between coronary anatomy lesions and changes in blood flow function. Specifically, it involves acquiring coronary function analysis data corresponding to the coronary CT angiography data, spatially registering the location of coronary anatomy lesions with the location of functional changes distributed along the coronary artery centerline based on the verified coronary lesion data and coronary function analysis data, and determining the coronary lesion pattern based on the positional correspondence and distribution status between the anatomy lesions and functional changes to obtain the coronary anatomy and function correspondence result data.
[0097] The coronary artery anatomy and function correspondence analysis includes the following steps:
[0098] Step S41: Coronary artery anatomical lesion localization, used to determine the reliable spatial range of coronary artery anatomical lesions, specifically, based on the verified coronary artery lesion data, mapping the analyzable vessel segment and the verified analyzable vessel segment to the corresponding coronary artery centerline, determining the start and end positions of each coronary artery lesion, and obtaining coronary artery anatomical lesion location data;
[0099] In practice, the coronary artery centerline obtained in step S1 is used as a unified spatial positioning reference. Cumulative arc length coordinates are established for each branch of the coronary artery centerline. The proximal starting point of the corresponding coronary artery centerline branch is taken as the zero point of the cumulative arc length. The cumulative arc length of each centerline sampling position is calculated sequentially along the distal direction of blood flow. Based on the spatial correspondence in the coronary vascular structure data, the coronary lesion areas in the analyzable vascular segment and the re-analyzable vascular segment are mapped to the corresponding coronary artery centerline branches, and the cumulative arc length position of the centerline corresponding to the cross section constituting each coronary lesion area is recorded.
[0100] Preferably, for a coronary artery lesion region that is continuous in the same space, the position of the minimum cumulative arc length on the corresponding coronary artery central line branch is determined as the starting position of the coronary artery lesion, and the position of the maximum cumulative arc length is determined as the ending position of the coronary artery lesion; for a coronary artery lesion region that is not continuous in space, the starting position and the ending position are determined separately; according to the verified coronary artery lesion data, the degree of luminal stenosis corresponding to each coronary artery lesion is read, and the coronary artery central line branch to which the coronary artery lesion belongs, the starting position, the ending position, and the degree of luminal stenosis are recorded together to obtain the anatomical lesion location data of the coronary artery;
[0101] Step S42: Localization of coronary artery function changes, used to determine the spatial range of coronary artery function abnormalities. Specifically, the coronary artery function analysis data is mapped to the corresponding coronary artery centerline. Based on the continuous changes of functional parameters along the coronary artery centerline, the local functional change area and the continuous functional change area are determined to obtain the location data of coronary artery function changes.
[0102] In specific implementation, coronary artery function analysis data corresponding to the coronary CT angiography data is acquired. The coronary artery function analysis data is continuous functional parameter data that has a spatial correspondence with the coronary artery centerline. In this embodiment, the functional parameters are obtained from the CT flow reserve obtained from the coronary CT angiography data, and the coronary artery centerline branch to which each functional parameter value belongs and the corresponding cumulative arc length position are recorded. According to the cumulative arc length coordinates established in step S41, the coronary artery function analysis data is mapped to the corresponding coronary artery centerline branch, and the functional parameters are unified to a centerline sampling interval of 1 mm through linear interpolation.
[0103] Preferably, for each coronary artery centralline branch, functional parameters are sequentially read from the first effective sampling position proximally towards the distal end. The current starting sampling position is used as the starting point of a candidate interval. When the cumulative decrease of the functional parameter relative to the starting point first reaches the functional change judgment value, the current sampling position is determined as the endpoint of the candidate interval. After determining a candidate interval, the search restarts from the next sampling position after the endpoint of the candidate interval to avoid generating a large number of overlapping candidate intervals due to the same functional change process. The functional change corresponding to the candidate interval is calculated according to the following formula:
[0104] ;
[0105] in, Indicates the position of the center line to The change in function between them is a dimensionless quantity; Indicates the position of the center line The corresponding functional parameter values are dimensionless. Indicates the starting position of the candidate interval, in millimeters; Indicates the position of the center line The corresponding functional parameter values are dimensionless. The terminator indicates the termination position of the candidate interval, in millimeters; the subscripts a and b represent the starting and ending sampling positions of the candidate interval, respectively; the functional change is used to characterize the cumulative decrease of the functional parameters within the corresponding centerline interval relative to the starting point of the interval;
[0106] In specific implementation, the functional change judgment value is determined based on the absolute value distribution of the differences in functional parameters obtained from repeated functional analysis of the same coronary artery centerline position in historical calibration samples. Preferably, the 95th percentile of the absolute value distribution is used, which is 0.05 in this embodiment. When the functional change amount of the candidate interval reaches the functional change judgment value and the corresponding centerline length is not greater than 20 mm, the candidate interval is determined as a local functional change region. When the corresponding centerline length is greater than 20 mm, and at least 80% of the sampling point pairs in all adjacent sampling point pairs within the candidate interval satisfy the condition that the functional parameter of the distal sampling position is not higher than the functional parameter of the corresponding proximal sampling position, the candidate interval is determined as a continuous functional change region. The 20 mm can be corrected using the 95th percentile of the length distribution of local functional change regions in historical calibration samples. The coronary artery centerline branch, start position, end position, region type, and corresponding functional change amount of the functional change region are recorded together to obtain coronary artery functional change location data.
[0107] Step S43: Anatomical function spatial matching, used to determine the correspondence between coronary artery anatomical lesions and coronary artery function changes. Specifically, based on the location data of the coronary artery anatomical lesions and the location data of the coronary artery function changes, spatial overlap analysis is performed on the anatomical lesion area and the functional change area. Anatomical lesions that meet the preset spatial correspondence conditions are associated with functional changes to obtain coronary artery anatomical function matching data.
[0108] In practice, anatomical lesion intervals are constructed based on the starting and ending positions in the coronary artery anatomical lesion location data, and functional change intervals are constructed based on the starting and ending positions in the coronary artery function change location data. Spatial matching is performed only on anatomical lesion intervals and functional change intervals belonging to the same coronary artery centralline branch, and both types of intervals are uniformly represented on the cumulative arc length coordinates of the coronary artery centralline branch. Considering the spatial positioning errors caused by coronary lesion boundary identification and functional parameter sampling, the starting position of the anatomical lesion interval is extended proximally by δ, and the ending position is extended distally by δ, forming an extended anatomical lesion interval. The extended range does not exceed the proximal start and distal end of the corresponding coronary artery centralline branch.
[0109] Preferably, the spatial coverage matching degree between the i-th coronary artery lesion and the j-th functional change region is calculated according to the following formula:
[0110] ;
[0111] in, The spatial coverage matching degree between the i-th coronary artery lesion and the j-th functional change region is represented by a dimensionless quantity. A function to calculate the length of the centerline interval, with the output unit being millimeters; The superscript indicates the extended anatomical lesion region corresponding to the i-th coronary artery lesion. This represents the spatial extension distance between the two ends of the anatomical lesion region, in millimeters. The subscript 'i' indicates the coronary artery lesion number. This represents the intersection of two centerline intervals; This represents the centerline interval corresponding to the j-th functional change area, where the subscript j indicates the functional change area number. This indicates that the shorter of the extended anatomical lesion interval length and the functional change interval length is taken; the spatial coverage matching degree is used to characterize the degree to which the shorter interval of the anatomical lesion interval and the functional change interval is covered by another spatial interval;
[0112] In practical implementation, the spatial expansion distance The location of coronary artery lesions with clear anatomical functional correspondences in historical calibration samples is determined based on the spatial differences between the start and end locations of the corresponding functional change areas. Preferably, the 95th percentile of the absolute value distribution of these spatial differences is used. In the absence of such historical calibration samples... Take 5 mm; when the spatial coverage matching degree reaches 0.50, establish an association between the corresponding coronary artery lesion and the functional change area; record the associated coronary artery central line branch, coronary artery lesion number, functional change area number and spatial coverage matching degree together to obtain coronary artery anatomy and function matching data;
[0113] Step S44: Coronary artery lesion pattern determination, used to distinguish the coronary artery lesion status corresponding to different anatomical-functional relationships. Specifically, based on the coronary artery anatomical-functional matching data, the vessel segment corresponding to the coronary artery lesion and local functional changes is identified as focal function-related lesion, the vessel segment where the degree of anatomical lesion is inconsistent with the degree of corresponding functional changes is identified as anatomical-functional inconsistent lesion, and the vessel corresponding to multiple consecutive coronary artery lesions and continuous functional changes is identified as diffuse functional impairment lesion, thus obtaining the coronary artery anatomical-functional corresponding result data.
[0114] In practice, the interrelated coronary artery lesions and functional change areas are read based on coronary anatomy and function matching data. The degree of anatomical lesion and the degree of functional change are determined according to the corresponding degree of luminal stenosis and the amount of functional change. In this embodiment, the 33rd and 67th percentiles of the distribution of luminal stenosis in historical calibration samples are used as the three-level division boundary for the degree of anatomical lesion, and the 33rd and 67th percentiles of the distribution of functional change are used as the three-level division boundary for the degree of functional change. The two are respectively divided into Level 1, Level 2, and Level 3 from low to high. The grading boundaries are fixed after being determined and used for subsequent determination of coronary artery lesion patterns.
[0115] As a further optimization of this embodiment, when the coronary lesion and the area of local functional change meet the spatial correspondence condition of step S43, and the absolute difference between the anatomical lesion severity level and the functional change severity level number does not exceed 1, the vessel segment where the corresponding coronary lesion is located is identified as a focal function-related lesion; when the absolute difference between the anatomical lesion severity level and the corresponding functional change severity level number is 2, the vessel segment where the corresponding coronary lesion is located is identified as an anatomical-functional inconsistency lesion.
[0116] In practice, when two or more coronary artery lesions within the same region of persistent functional change are associated with the same central branch of the same coronary artery, and arranged in order from proximal to distal, if the central line interval between the termination position of the preceding coronary artery lesion and the starting position of the following coronary artery lesion does not exceed the continuous lesion interval determination value, the corresponding central branch of the coronary artery is identified as a diffuse functional impairment lesion. The continuous lesion interval determination value is preferably the 95th percentile of the distribution of central line intervals of adjacent coronary artery lesions corresponding to the same region of persistent functional change in historical calibration samples, and 10 mm is used when the historical calibration samples are unavailable.
[0117] As a further optimization of this implementation, when the same blood vessel simultaneously meets two or more coronary artery lesion pattern determination conditions, the coronary artery lesion patterns that are met are recorded separately, and one coronary artery lesion pattern is not used to cover other coronary artery lesion patterns that have already met the determination conditions; for coronary artery anatomy and function matching relationships that do not meet the above three types of coronary artery lesion pattern determination conditions, the corresponding coronary artery anatomy and function matching data are retained without assigning a coronary artery lesion pattern; the coronary artery lesion pattern, the corresponding coronary artery lesion spatial range and its associated functional change area are recorded together to obtain the corresponding coronary artery anatomy and function result data.
[0118] By performing the above operations, this solution addresses the inconsistency between the degree of coronary artery anatomical lesions and the actual impact on blood flow function. It employs an anatomical-functional spatial matching technique based on the coronary artery centerline. This involves spatially matching reliable coronary artery anatomical lesions with corresponding functional changes, and differentiating different coronary lesion patterns based on the local or continuous distribution of functional changes. This enables the differentiation and identification of focal function-related lesions, anatomically-functionally inconsistent lesions, and diffuse functional impairment lesions. For example, when multiple consecutive coronary artery lesions correspond to continuous functional changes along the vessel, they are identified as diffuse functional impairment lesions, rather than solely relying on the maximum local stenosis for risk assessment.
[0119] Example 6, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S5, the coronary artery risk prediction for the elderly is used to determine the coronary artery risk of the elderly test subjects based on reliable coronary artery lesions and their functional impact. Specifically, it involves obtaining the basic clinical data of the elderly test subjects, constructing coronary artery risk prediction features based on the basic clinical data, the reviewed coronary artery lesion data, and the corresponding results of coronary artery anatomy and function, and inputting them into a pre-trained artificial intelligence risk prediction model to perform coronary artery risk prediction, thereby obtaining the coronary artery risk prediction result data for the elderly.
[0120] In specific implementation, basic clinical data of elderly subjects are obtained. In this embodiment, the elderly subjects are those who are 60 years old or older on the day of the coronary CT angiography examination. The basic clinical data include age, systolic blood pressure, low-density lipoprotein cholesterol, and diabetes status. Among them, age is the actual age in years on the day of examination, systolic blood pressure is the average of three consecutive measurements under resting conditions before the examination, low-density lipoprotein cholesterol is the most recent laboratory test result within 30 days before the examination, and diabetes status is recorded as 1 according to the clear diagnosis result in the clinical medical record at the time of examination, and 0 if there is no clear diagnosis of diabetes.
[0121] In practice, based on the branching affiliation of the coronary artery centerline, the continuous vessel segments determined in step S1 are uniformly assigned to the left main coronary artery region, the left anterior descending artery region, the left circumflex artery region, and the right coronary artery region. The left main coronary artery region includes the left main coronary artery and its branches that originate directly from it and cannot be assigned to the other three coronary artery regions; the left anterior descending artery region includes the left anterior descending artery and its subordinate branches; the left circumflex artery region includes the left circumflex artery and its subordinate branches; and the right coronary artery region includes the right coronary artery and its subordinate branches. For branch vessels, the corresponding coronary artery region is determined by tracing proximally along the coronary artery centerline and based on their superior main coronary artery affiliation, thus ensuring that each continuous vessel segment has a unique coronary artery region affiliation.
[0122] Preferably, for each coronary artery region, the analyzable vessel segment and the corresponding luminal stenosis degree of the re-analyzable vessel segment are read from the re-verified coronary artery lesion data. For analyzable or re-analyzable vessel segments that meet the conditions for quantifying coronary artery lesions but have not identified any coronary artery lesions, the corresponding luminal stenosis degree is recorded as 0. When an analyzable or re-analyzable vessel segment exists in the corresponding coronary artery region, the maximum value of the luminal stenosis degree of the analyzable or re-analyzable vessel segment is used as the maximum luminal stenosis degree of the coronary artery region. Unquantifiable vessel segments are not included in the calculation of the maximum luminal stenosis degree. When no analyzable or re-analyzable vessel segment exists in the corresponding coronary artery region, the maximum luminal stenosis degree is recorded as a missing value.
[0123] In practice, based on the corresponding data of coronary artery anatomy and function, three states are set for each coronary artery region: focal function-related lesion state, anatomical-functional inconsistency lesion state, and diffuse functional impairment lesion state. When a corresponding coronary artery lesion pattern exists in the corresponding coronary artery region, the corresponding state is recorded as 1; otherwise, it is recorded as 0. Two or more coronary artery lesion patterns can exist simultaneously in the same coronary artery region. At the same time, an unquantifiable state is set based on the reviewed coronary artery lesion data. When one or more unquantifiable vessel segments exist in the corresponding coronary artery region, it is recorded as 1; otherwise, it is recorded as 0.
[0124] As a further optimization of this embodiment, each coronary artery region is configured with 5-dimensional coronary artery region features in a fixed order of maximum luminal stenosis, focal functionally related lesion status, anatomically and functionally inconsistent lesion status, diffuse functional impairment lesion status, and unquantifiable status. The features of four coronary artery regions are then connected in a fixed order of left main coronary artery region, left anterior descending artery region, left circumflex artery region, and right coronary artery region to form 20-dimensional coronary artery imaging and functional features. Finally, four basic clinical data points are connected in the order of age, systolic blood pressure, low-density lipoprotein cholesterol, and diabetes status to form 24-dimensional coronary artery risk prediction features. The order of these features remains consistent during the training and prediction process of the artificial intelligence risk prediction model.
[0125] Preferably, a sample of elderly subjects with complete basic clinical data and corresponding follow-up is obtained, and the elderly subjects are divided into training, validation, and test sets in a 7:2:1 ratio, so that the data of the same elderly subject is included in only one subset of data; whether myocardial infarction or unplanned coronary revascularization due to coronary artery disease occurred within 24 months after coronary CT angiography is used as a training label. The elderly subjects with any of the aforementioned coronary artery-related events are marked as positive, and the elderly subjects who did not have the aforementioned coronary artery-related events within 24 months and completed the 24-month follow-up are marked as negative.
[0126] In specific implementation, the historical elderly test subject samples in the training set, validation set and test set are processed according to the judgment limits and grading limits determined in steps S2 to S4, and the corresponding 24-dimensional coronary artery risk prediction features are constructed according to the same method described above; the judgment limits and grading limits remain consistent in the training, validation, testing and subsequent coronary artery risk prediction of the artificial intelligence risk prediction model for elderly test subjects.
[0127] Preferably, the artificial intelligence risk prediction model adopts the LightGBM binary classification model, and the model is trained using the 24-dimensional coronary artery risk prediction features and training labels corresponding to the training set. The LightGBM binary classification model adopts a binary classification objective function, uses the binary log loss as the training loss evaluation index, sets the learning rate to 0.05, the maximum tree depth to 6, the number of leaf nodes to 31, and the maximum number of iterations to 500. During the training process, the model is selected according to the binary log loss corresponding to the validation set. When the binary log loss of the validation set does not decrease for 30 consecutive iterations, the training stops, and the model parameters corresponding to the lowest binary log loss of the validation set are retained.
[0128] Preferably, for the missing values of the maximum luminal stenosis in the 24-dimensional coronary artery risk prediction features, the missing state is retained and processed using the missing value branching mechanism of the LightGBM model. At the same time, combined with the unquantifiable state corresponding to the same coronary artery region, the artificial intelligence risk prediction model can distinguish between the state of no clear coronary artery lesion with a maximum luminal stenosis of 0 and the state of lacking reliable stenosis due to the unquantifiable nature of the vessel segment. The training set, validation set, test set and subsequent prediction data all use the same missing value expression method, and the missing values are not filled with zero values.
[0129] In practice, the 24-dimensional coronary artery risk prediction features corresponding to the elderly test subjects are input into the trained artificial intelligence risk prediction model in the same feature order as during the model training phase. The model outputs a coronary artery risk prediction probability in the range of 0 to 1. The coronary artery risk prediction probability represents the predicted probability that the elderly test subject will experience the coronary artery-related event within 24 months after the coronary artery CT angiography examination. The coronary artery risk prediction probability, the spatial location of coronary artery lesions and the information on unquantifiable vessel segments in the reviewed coronary artery lesion data, and the coronary artery lesion pattern in the corresponding coronary artery anatomy function data are recorded together to obtain the coronary artery risk prediction result data for the elderly.
[0130] Example 7, see Figure 1 and Figure 2 Based on the above embodiments, this embodiment provides an artificial intelligence-based coronary artery risk prediction system for the elderly, including a coronary artery structure extraction module, a vessel segment interpretation module, a vessel segment verification module, an anatomical function correspondence module, and a coronary artery risk prediction module.
[0131] The coronary artery structure extraction module is used to extract the coronary artery centerline, lumen boundary and plaque region from the coronary artery CT angiography data of elderly subjects, establish the spatial correspondence between continuous vessel segments and coronary artery CT angiography data, obtain coronary artery structure data, and send the coronary artery structure data to the vessel segment interpretation module.
[0132] The vascular segment interpretation module is used to identify the influence of calcification boundaries, abnormal lumen structure and local imaging abnormalities based on the coronary artery structure data, determine the lesion analysis path of each continuous vascular segment, obtain the coronary artery segment interpretation result data, and send the coronary artery segment interpretation result data to the vascular segment verification module.
[0133] The vessel segment verification module is used to call the corresponding alternative reconstruction sequence to re-identify the lumen and verify consistency based on the readable result data of the coronary vessel segment, to obtain the verified coronary lesion data, and to send the verified coronary lesion data to the anatomical function corresponding module and the coronary risk prediction module.
[0134] The anatomical function corresponding module is used to spatially match the location of coronary anatomical lesions with the location of coronary function changes based on the reviewed coronary lesion data and coronary function analysis data, determine the corresponding coronary lesion pattern, obtain the coronary anatomical function corresponding result data, and send the coronary anatomical function corresponding result data to the coronary risk prediction module.
[0135] The coronary artery risk prediction module is used to construct coronary artery risk prediction features based on the basic clinical data, reviewed coronary artery lesion data, and corresponding results of coronary artery anatomy and function of the elderly test subjects. It then uses a pre-trained artificial intelligence risk prediction model to predict coronary artery risk, obtains coronary artery risk prediction result data for the elderly, and outputs the coronary artery risk prediction result data for the elderly.
[0136] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0137] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.
[0138] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. An artificial intelligence-based method for predicting coronary artery risk in the elderly, characterized in that: The method includes the following steps: Step S1: Extract coronary artery structure data to obtain coronary artery structure data; Step S2: Coronary artery segment readability identification. Based on the coronary artery structure data, calcification boundary influence identification, continuous lumen structure verification, and local imaging anomaly identification are performed on each continuous segment. Based on the identification results, the lesion analysis path for the corresponding continuous segment is determined to obtain the coronary artery segment readability result data. This includes the following steps: Step S21: Calcification boundary influence identification; Step S22: Continuous lumen structure verification; Step S23: Local imaging anomaly identification; Step S24: Segment readability path determination. Step S3: Review the interfered vessel segment, determine the type of image interference corresponding to the vessel segment to be reviewed, select an alternative reconstruction sequence from the reconstruction data corresponding to the same coronary CT angiography examination, re-identify the lumen boundary of the vessel segment to be reviewed in the alternative reconstruction sequence, and perform consistency verification on the lumen identification results in different available reconstruction sequences to obtain the reviewed coronary lesion data. Step S4: Coronary artery anatomy and function correspondence analysis. Based on the verified coronary artery lesion data and coronary artery function analysis data, spatial registration is performed between the location of the coronary artery anatomy lesion and the location of the functional changes distributed along the coronary artery centerline. The coronary artery lesion pattern is determined based on the positional correspondence and distribution of the anatomy lesion and functional changes, and the coronary artery anatomy and function correspondence result data is obtained. Step S5: Predict coronary artery risk in the elderly, and obtain the prediction results data for coronary artery risk in the elderly.
2. The artificial intelligence-based method for predicting coronary artery risk in the elderly according to claim 1, characterized in that: In step S1, the coronary artery structure data is extracted by obtaining the coronary artery CT angiography data of the elderly test subjects. The coronary artery centerline, lumen boundary and plaque region are extracted by artificial intelligence image recognition model. The coronary artery is divided into continuous vessel segments according to the coronary artery centerline. The spatial correspondence between each continuous vessel segment and the coronary artery CT angiography data is established to obtain the coronary artery structure data.
3. The artificial intelligence-based method for predicting coronary artery risk in the elderly according to claim 2, characterized in that: In step S2, the calcification boundary influence identification involves extracting cross-sections corresponding to continuous vessel segments along the coronary artery centerline, identifying the location where the calcification region covers or interferes with the lumen boundary based on the positional relationship between the calcification region and the lumen boundary, and obtaining calcification boundary influence data. The continuous lumen structure verification involves comparing the lumen position, lumen area, and boundary morphology of adjacent cross sections based on the calcification boundary influence data, identifying local abrupt changes that are inconsistent with adjacent lumen structures, and obtaining abnormal lumen structure data.
4. The artificial intelligence-based method for predicting coronary artery risk in the elderly according to claim 3, characterized in that: In step S2, the local imaging anomaly identification involves verifying the continuity of the centerline, the lumen edge, and the angiographic filling status of continuous vessel segments based on the coronary vessel structure data and lumen structure anomaly data, to determine the location of vessels with motion interference or angiographic filling abnormalities, thereby obtaining local imaging anomaly data. The vascular segment interpretation path is determined by analyzing continuous vascular segments that meet the conditions for quantifying coronary artery lesions, based on the calcification boundary influence data, lumen structure abnormality data, and local imaging abnormality data. Continuous vascular segments with luminal structure abnormalities or localized missing lumen boundaries affected by image interference and with corresponding alternative reconstruction data are identified as vascular segments to be reviewed. Continuous vascular segments that do not meet the conditions for quantifying coronary artery lesions are identified as unquantifiable vascular segments, thus obtaining interpretable coronary vascular segment data.
5. The artificial intelligence-based method for predicting coronary artery risk in the elderly according to claim 4, characterized in that: In step S3, the interference vessel segment review involves verifying the consistency of lumen identification results in different available reconstruction sequences. Vessel segments that meet the consistency requirements are identified as analyzable vessel segments for review, while those that do not meet the consistency requirements are identified as unquantifiable vessel segments, thus obtaining the reviewed coronary lesion data.
6. The artificial intelligence-based method for predicting coronary artery risk in the elderly according to claim 5, characterized in that: In step S4, the coronary artery anatomy function correspondence analysis includes the following steps: Step S41: Coronary artery anatomical lesion localization. Based on the verified coronary artery lesion data, the analyzable vessel segment and the verified analyzable vessel segment are mapped to the corresponding coronary artery centerline to determine the starting and ending positions of each coronary artery lesion and obtain the coronary artery anatomical lesion location data. Step S42: Locating coronary artery function changes. Map the coronary artery function analysis data to the corresponding coronary artery centerline. Determine the local functional change area and the continuous functional change area based on the continuous changes of functional parameters along the coronary artery centerline to obtain the location data of coronary artery function changes. Step S43: Anatomical function spatial matching. Based on the coronary artery anatomical lesion location data and coronary artery function change location data, perform spatial overlap analysis on the anatomical lesion area and the functional change area, establish association between anatomical lesions and functional changes that meet the preset spatial correspondence conditions, and obtain coronary artery anatomical function matching data. Step S44: Coronary artery lesion pattern determination. Based on the coronary artery anatomy and function matching data, the vessel segments corresponding to local anatomical lesions and local functional changes are identified as focal function-related lesions. The vessel segments where the degree of anatomical lesion and the degree of corresponding functional changes are inconsistent are identified as anatomical-functional inconsistent lesions. Multiple consecutive coronary artery lesions and vessels corresponding to continuous functional changes are identified as diffuse functional impairment lesions. Coronary artery anatomy and function corresponding result data are obtained.
7. The artificial intelligence-based method for predicting coronary artery risk in the elderly according to claim 6, characterized in that: In step S5, the elderly coronary artery risk prediction involves acquiring basic clinical data of the elderly test subjects, constructing coronary artery risk prediction features based on the basic clinical data, the reviewed coronary artery lesion data, and the corresponding results of coronary artery anatomy and function, and inputting them into a pre-trained artificial intelligence risk prediction model to perform coronary artery risk prediction, thereby obtaining the elderly coronary artery risk prediction result data.
8. An artificial intelligence-based coronary artery risk prediction system for the elderly, used to implement the artificial intelligence-based coronary artery risk prediction method for the elderly as described in any one of claims 1-7, characterized in that: It includes a coronary artery structure extraction module, a vessel segment interpretation module, a vessel segment verification module, an anatomical function correspondence module, and a coronary artery risk prediction module.
9. The artificial intelligence-based coronary artery risk prediction system for the elderly according to claim 8, characterized in that: The coronary artery structure extraction module is used to extract coronary artery structure data, obtain coronary artery structure data, and send the coronary artery structure data to the vessel segment interpretation module. The readable segment module is used for coronary artery segment readableness identification, obtaining coronary artery segment readableness result data, and sending the coronary artery segment readableness result data to the vascular segment verification module; The vessel segment verification module is used to verify the interfered vessel segment, obtain the verified coronary artery lesion data, and send the verified coronary artery lesion data to the anatomical function corresponding module and the coronary artery risk prediction module. The anatomical function correspondence module is used for coronary artery anatomical function correspondence analysis, to obtain coronary artery anatomical function correspondence result data, and to send the coronary artery anatomical function correspondence result data to the coronary artery risk prediction module; The coronary artery risk prediction module is used for predicting coronary artery risk in the elderly, obtaining coronary artery risk prediction result data for the elderly, and outputting the coronary artery risk prediction result data for the elderly.
