Airway CT image feature extraction method for pre-anesthesia risk assessment

Abstract

The present application relates to the technical field of image feature extraction, and particularly relates to an airway CT image feature extraction method for pre-anesthesia risk assessment. The present application first obtains the pulling transmission resistance coefficients between different muscle regions according to the spatial dislocation relationship and the projection overlap relationship between different muscle regions between adjacent anatomical layers of airway CT images, and matches the different muscle regions based on the pulling transmission resistance coefficients to obtain the hierarchical transmission loss coefficients of each anatomical layer. Then, according to the change characteristics of the hierarchical transmission loss coefficients of the anatomical layer in the preset pulling stress axis, the airway risk feature parameters of the pre-anesthesia patient are extracted. The present application analyzes the muscle bundle distribution and force transmission impedance by means of the spatial dislocation and overlap characteristics of the interlayer muscle regions, simulates stress transmission through interlayer muscle region matching to analyze the hierarchical transmission loss of the anatomical layer, and then extracts the airway risk features to assist pre-anesthesia risk assessment.

Description

Technical Field

[0001] This invention relates to the field of image feature extraction technology, and more specifically to a method for extracting features from airway CT images for pre-anesthesia risk assessment. Background Technology

[0002] Maintaining an open upper airway is fundamental to ensuring safe ventilation in patients during the induction of general anesthesia. In obese patients or those with obstructive sleep apnea, heterogeneous fatty infiltration is often present within the tongue, a pathological change that breaks down dense muscle fiber bundles into fragmented, diffuse, independent clumps. While neuromuscular compensation mechanisms can maintain airway tone in the conscious state, after muscle relaxation due to anesthesia induction, the fragmented tissue structures cannot effectively counteract gravity and negative pressure, leading to severe hidden posterior displacement of the tongue and airway obstruction.

[0003] Current airway assessment techniques largely rely on surface examinations such as the Mallampati classification or routine CT imaging. Surface examinations cannot provide insight into internal structures, while CT imaging primarily focuses on macroscopic morphological features such as the cross-sectional area of ​​the airway lumen. However, for the same airway cross-sectional area, the mechanical conduction properties of intact muscle differ significantly from those of muscle infiltrated by fat. This makes it difficult to quantify mechanical conduction failure caused by muscle bundle fragmentation and disjointed connections, and to distinguish between "anatomical stenosis" and "functional structural loosening." Consequently, relying solely on airway morphological features is insufficient to accurately characterize the risk of airway collapse in a state of muscle relaxation. Summary of the Invention

[0004] To address the technical problem that fatty infiltration makes it difficult to accurately characterize the risk of airway collapse under muscle relaxation conditions using airway morphology features, the present invention aims to provide a method for extracting airway CT image features for pre-anesthesia risk assessment. The specific technical solution adopted is as follows: Acquire CT images of the airway at each anatomical layer along the pre-set traction force axis before anesthesia, and acquire all muscle regions in each airway CT image; wherein, the pre-set traction force axis is the direction of the line connecting the mental spine of the mandible to the posterior wall of the airway; Between airway CT images of adjacent anatomical layers, the traction transmission resistance coefficient between different muscle regions is obtained based on the spatial misalignment and projection overlap relationship between the different muscle regions. Based on the traction transmission resistance coefficient, inter-layer matching is performed on different muscle regions to obtain the layer transmission loss coefficient of each anatomical layer. Based on the variation characteristics of the hierarchical transmission loss coefficient of the anatomical layer along the preset traction force axis, airway risk characteristic parameters of the patient before anesthesia are extracted.

[0005] Furthermore, the method for obtaining the muscle region includes: Threshold segmentation was performed on the airway CT images of each anatomical layer to obtain all muscle pixels, and region connectivity detection was performed on the muscle pixels to obtain all muscle regions.

[0006] Furthermore, the method for obtaining the tensile resistance coefficient includes: Between adjacent anatomical layers, based on the spatial distance and grayscale change information between the geometric centers corresponding to different muscle regions, as well as the overlapping area of ​​connected regions, the basic traction transmission resistance parameters between different muscle regions are obtained. For each muscle region, a distributed transfer penalty weight is obtained based on the distribution characteristics of the basic traction transfer resistance parameters between the muscle region and the airway CT images of adjacent anatomical layers. The basic traction transmission resistance parameter is weighted using the dispersed transmission penalty weight to obtain the traction transmission resistance coefficient between each muscle region and the airway CT image of the adjacent anatomical layer.

[0007] Furthermore, the method for obtaining the basic tensile resistance parameters includes: Between different muscle regions, the normalized result of the Euclidean distance between the corresponding geometric centers is used as the first resistance parameter, the negative correlation normalized result of the pixel mean of the voxel of the path connecting the corresponding geometric centers is used as the second resistance parameter, and the negative correlation normalized result of the Jaccard similarity coefficient between the corresponding pixel sets is used as the third resistance parameter. The first resistance parameter, the second resistance parameter, and the third resistance parameter are weighted and fused to obtain the basic traction transmission resistance parameters between different muscle regions.

[0008] Furthermore, the method for obtaining the distributed propagation penalty weights includes: Between different muscle regions, the negative correlation normalization result of the basic traction transmission resistance parameter is used as the interlayer muscle connection probability; for each muscle region, the normalization result of the entropy of the interlayer muscle connection probability between it and each muscle region in the airway CT image of the adjacent anatomical layer is added with a constant 1 and used as the dispersion transmission penalty weight.

[0009] Furthermore, the interlayer matching of different muscle regions based on the traction transmission resistance coefficient includes: Between adjacent anatomical layers, based on the bipartite graph matching algorithm and the traction transmission resistance coefficient between different muscle regions, all matching pairs between all different muscle regions are obtained.

[0010] Furthermore, bipartite graph matching algorithms include at least the Hungarian algorithm.

[0011] Furthermore, the method for obtaining the hierarchical transmission loss coefficient includes: In the airway CT image of each anatomical layer, the normalized result of the area of ​​each muscle region is used as the transfer weight. The transfer weight is used to weight the traction transfer resistance coefficient between the matching pairs to which the muscle region belongs. The weighted sum is used as the layer transfer loss coefficient of the corresponding anatomical layer.

[0012] Furthermore, the method for obtaining the airway risk characteristic parameters includes: An axial transmission loss curve is fitted based on the hierarchical transmission loss coefficient of each anatomical layer; the maximum value in the axial transmission loss curve is taken as the maximum airway rupture risk value, and the integral of the axial transmission loss curve is taken as the cumulative airway looseness value; the maximum airway rupture risk value and the cumulative airway looseness value are taken as airway risk characteristic parameters.

[0013] Furthermore, the method for obtaining the axial transmission loss curve includes: The hierarchical transmission loss coefficients of each anatomical layer are sorted in order along the preset tensile force axis and then smoothed to obtain the axial transmission loss curve.

[0014] The present invention has the following beneficial effects: This invention first acquires CT images of the airway at each anatomical layer along the pre-set traction force axis before anesthesia, and obtains all muscle regions in each airway CT image, providing a basis for subsequent assessment of axial transmission of muscle mechanics. Then, between airway CT images of adjacent anatomical layers, based on the spatial misalignment and projection overlap between different muscle regions, it analyzes the axial transmission resistance of muscle bundles and interlayer slippage, obtaining the traction transmission resistance coefficient between different muscle regions. Based on the traction transmission resistance coefficient, it performs interlayer matching of different muscle regions to simulate the force transmission pattern of biological tissue, thereby analyzing the interlayer matching transmission to obtain the hierarchical transmission loss coefficient of each anatomical layer. Furthermore, based on the variation characteristics of the hierarchical transmission loss coefficient of the anatomical layer along the pre-set traction force axis, it analyzes the axial transmission risk, and then extracts the airway risk characteristic parameters of the patient before anesthesia. This invention utilizes the spatial misalignment and overlap characteristics of interlayer muscle regions to analyze muscle bundle distribution and force transmission impedance, and simulates stress transmission through interlayer muscle region matching to analyze hierarchical transmission loss of anatomical layers, thereby extracting airway risk characteristics to assist in pre-anesthesia risk assessment. Attached Figure Description

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

[0016] Figure 1 A flowchart of a method for extracting airway CT image features for preanesthesia risk assessment, provided as an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for obtaining the tensile resistance coefficient according to an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for airway CT image feature extraction for pre-anesthesia risk assessment proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following describes in detail, with reference to the accompanying drawings, a specific scheme for airway CT image feature extraction for pre-anesthesia risk assessment provided by the present invention.

[0020] Please see Figure 1 The document illustrates a flowchart of a method for extracting airway CT image features for pre-anesthesia risk assessment, provided by an embodiment of the present invention, specifically including: Step S1: Obtain CT images of the airway in each anatomical layer along the preset traction force axis before anesthesia, and obtain all muscle regions in each airway CT image; wherein, the preset traction force axis is the direction of the line connecting the mental spine of the mandible to the posterior wall of the airway.

[0021] In one embodiment of the present invention, the raw DICOM data sequence from a patient's neck CT scan is first read. Using an anatomical landmark recognition algorithm or a manual interaction method, two key anatomical points are located in three-dimensional voxel space: Mechanical starting point: the center point of the mental spine of the mandible, the bony attachment point of the genioglossus muscle, and also the main force point to prevent the tongue from falling back; Mechanical endpoint: The soft tissue boundary point at the level of the junction of the posterior airway wall and the soft palate on the midsagittal plane. This point represents the farthest point that the soft tissue may come into contact with when the airway collapses. The line connecting the mental spine of the mandible (mechanical starting point) to the posterior wall of the airway (mechanical ending point) is taken as the preset traction force axis, and a right-handed Cartesian coordinate system is constructed with the mechanical starting point as the origin. The preset traction force axis (w-axis) is defined as the principal force axis, and the positive direction of the principal force axis is from the mechanical starting point to the mechanical ending point. This axis represents the main force direction of muscle contraction against airway collapse (i.e., the equipotential line of the mandibular lifting operation). The horizontal axis (u-axis) is perpendicular to the w-axis plane and corresponds to the left-right direction in anatomy. The vertical axis (v-axis) is perpendicular to the w-axis plane and corresponds to the up-down direction in anatomy (approximately the cephalopod axis direction).

[0022] Further, with the main force axis as the axial center line, a circular area with a radius of 15mm-25mm is set for the slice sampling range; in this embodiment, 15mm is used. This circular area covers the core antigravity muscle group, including the geniohyoid muscle and the geniohyoid muscularis, while excluding irrelevant posterior neck muscles and the spine. The original CT scan data is resampled in three-dimensional voxel space using a trilinear interpolation algorithm. The size of the resampled voxels is set to 1mm×1mm×1mm, thereby obtaining a series of sampling slices in the preset traction force axial direction. The axial stacking result of all sampling slices resembles a cylinder. Each sampling slice corresponds to an airway CT image on an anatomical layer with a thickness of 1mm.

[0023] In other embodiments, the implementer may also adjust the slice sampling range and resampling size as needed. The trilinear interpolation algorithm and resampling are well-known techniques and will not be described in detail here.

[0024] Severe fatty infiltration breaks down the originally continuous muscle fiber bundles into fragmented and diffuse structures. Only spatially continuous independent muscle blocks can serve as effective force transmission units, leading to a decline in the tongue's biomechanical transmission performance. This results in the inability to maintain airway tension as normally, potentially causing airway obstruction after anesthesia and muscle relaxation. Therefore, embodiments of this invention further identify muscles in each anatomical layer, i.e., acquiring all muscle regions in each airway CT image, providing a basis for subsequent assessment of muscle biomechanical transmission.

[0025] Preferably, in one embodiment of the present invention, considering that the CT value of muscle is significantly higher than that of fat and air, and its grayscale value is brighter on airway CT images, the method for obtaining the muscle region includes: Threshold segmentation was performed on the airway CT images of each anatomical layer to obtain all muscle pixels, and region connectivity detection was performed on the muscle pixels to obtain all muscle regions.

[0026] Specifically, the preferred CT value threshold range for muscles is set to +40HU to +80HU. Pixels in the airway CT image with CT values ​​within this threshold range are marked as foreground, i.e., muscle pixels, while the rest are marked as background (fat, air, or other tissue regions). Furthermore, an 8-neighborhood connected component labeling algorithm is executed to merge adjacent muscle pixels into a single muscle region.

[0027] It should be noted that the application of the 8-neighborhood connected component labeling algorithm is a well-known technology and will not be elaborated further; in other embodiments, implementers may also use other adaptive threshold segmentation algorithms or deep learning-based labeling algorithms to label muscle pixels.

[0028] Step S2: Between airway CT images of adjacent anatomical layers, based on the spatial misalignment and projection overlap between different muscle regions, obtain the traction transmission resistance coefficient between different muscle regions, and perform inter-layer matching between different muscle regions based on the traction transmission resistance coefficient to obtain the layer transmission loss coefficient of each anatomical layer.

[0029] In continuum mechanics, the transmission of tensile stress depends on the material continuity of the medium in the direction of force. If the physical contact area between muscle fibers in adjacent layers decreases, it means that the effective stress transmission channel is narrowed. Fat infiltration causes the muscle bundles to break down in space, so that the stress of the upper muscle area cannot be completely covered to the lower muscle area. Furthermore, since ideal anti-collapse muscles should be arranged in a straight line along the preset traction force axis, the muscle contraction force is mainly converted into axial tension to resist collapse. When the center of the muscle bundles of adjacent anatomical layers shifts laterally, i.e., spatial misalignment, the axial tension is decomposed into a normal component perpendicular to the contact surface and a tangential component (shear force) parallel to the contact surface. Soft tissues are much weaker in resisting shear force than in resisting tensile force, and are very prone to interlayer slippage at the misalignment point, resulting in a sharp decrease in force transmission efficiency. Since the overlapping and spatial misalignment of muscle regions between layers can reflect traction transmission resistance, this embodiment of the invention obtains the traction transmission resistance coefficient between different muscle regions based on the spatial misalignment and projection overlap between airway CT images of adjacent anatomical layers. The traction transmission resistance coefficient quantifies the mechanical impedance of the interlayer muscles, preparing for subsequent assessment of airway obstruction risk characteristics after anesthesia and muscle relaxation.

[0030] It should be noted that the analysis method for the traction transmission resistance parameters between different muscle regions of adjacent anatomical layers is consistent. In this embodiment, any muscle region in any anatomical layer K is taken as the target region, and any muscle region in the next adjacent anatomical layer K+1 in the positive direction of the preset traction force axis is taken as the reference region. The traction transmission resistance coefficient between the target region and the reference region is used as an example for analysis and description, and will not be repeated one by one.

[0031] It should be noted that when there is no muscle region in the anatomical layer, the layer transmission loss coefficient of the anatomical layer is directly set to the preset maximum blocking threshold, such as 2, and the subsequent S2 analysis and calculation are not performed.

[0032] Preferably, in one embodiment of the present invention, please refer to Figure 2 The diagram illustrates a flowchart of a method for obtaining the tensile resistance coefficient according to an embodiment of the present invention, specifically including: Step S201: Between adjacent anatomical layers, based on the spatial distance and grayscale change information between the geometric centers of different muscle regions, as well as the overlapping area of ​​connected regions, the basic traction transmission resistance parameters between different muscle regions are obtained.

[0033] Since the spatial distance between the geometric centers of the interlayer muscle regions can help assess spatial misalignment, and thus characterize the efficiency of the conduction path or shear loss; Furthermore, since the grayscale change information on the path connecting the geometric centers of the interlayer muscle regions can characterize the properties of the interlayer transmission medium and the risk of breakage, when the connection path is filled with low-grayscale adipose tissue, it indicates that the muscle bundle has undergone substantial biological breakage or severe infiltration, and the traction transmission resistance is relatively greater. Furthermore, the overlapping area of ​​connected domains between interlayer muscle regions can help assess the continuity of conduction channel information and muscle bundle contact, and the Jaccard similarity coefficient can help assess overlapping information by using the idea of ​​connected domain pixel sets, which can in turn help quantify the resistance to traction transmission. Based on this, in a preferred embodiment of the present invention, the method for obtaining the basic tensile resistance parameters includes: Between different muscle regions, the normalized result of the Euclidean distance between the corresponding geometric centers is used as the first resistance parameter, the negative correlation normalized result of the pixel mean of the voxel of the path connecting the corresponding geometric centers is used as the second resistance parameter, and the negative correlation normalized result of the Jaccard similarity coefficient between the corresponding pixel sets is used as the third resistance parameter. The first resistance parameter, the second resistance parameter, and the third resistance parameter are weighted and fused to obtain the basic stretching resistance parameters between different muscle regions.

[0034] As an example, a (two-dimensional) image coordinate system is first constructed with the center of each airway CT image as the origin, and then the coordinates of each pixel in each airway CT image are determined; the centroid of each muscle region is taken as the geometric center and the centroid coordinates are determined; in other examples, the implementer may also construct the image coordinate system by themselves, or use the centroid or centroid as the geometric center. The Euclidean distance D between the centroid coordinates of the target area and the reference area is normalized to obtain the first resistance parameter. The normalization method is to divide the Euclidean distance D by the maximum Euclidean distance between the centroid coordinates of different muscle regions between adjacent anatomical layers. The implementer can also divide by a preset maximum distance, such as the diameter of the slice sampling range, which is 30 mm. The second resistance parameter is obtained by negatively normalizing the pixel (CT value) values ​​of the voxels along the connection path between the target region and the reference region in 3D voxel space, based on the corresponding centroids. The negative correlation normalization is performed using a piecewise linear function: for example, when the pixel mean is ≥ +60HU, it is determined to be a typical muscle connection path, and the second resistance parameter is set to 0; when the pixel mean is ≤ -100HU, it is determined to be a typical fat connection path, and the second resistance parameter is set to 1; when the pixel mean is between -100HU and +60HU, the pixel mean is used as the independent variable x and mapped to... In this process, the mapping result is used as the second resistance parameter, with a value range of (0,1). The pixel coordinate sets of the target area and the reference area are counted separately, and then the Jaccard similarity coefficient J of the two pixel coordinate sets is calculated. Since the value of J is in the range of 0-1, J is further negatively normalized by 1-J to obtain the third drag parameter. Then, the first resistance parameter, the second resistance parameter, and the third resistance parameter are weighted and summed with weights of 0.3, 0.4, and 0.3 respectively. The weighted summation result is used as the basic traction transfer resistance parameter between the target area and the reference area.

[0035] It should be noted that in other embodiments, implementers may also use other normalization or negative correlation normalization methods, and may also adjust the weighting values ​​themselves.

[0036] By changing the target area and the reference area, the basic traction transmission resistance parameters between different muscle regions in all adjacent anatomical layers can be obtained.

[0037] Step S202: For each muscle region, obtain the dispersion transfer penalty weight based on the distribution characteristics of the basic traction transfer resistance parameters between it and each muscle region in the airway CT image of the adjacent anatomical layer.

[0038] Because severe fatty infiltration can cause muscle bundles to break down into diffusely distributed fragments, this change in topology means that the muscle region in the upper anatomical layer (with the pre-set traction force axis) may face multiple dispersed connected muscle regions in the next adjacent anatomical layer. That is, the upstream muscle bundle faces multiple dispersed downstream muscle bundles, which in turn causes the traction force to be unable to be concentrated and transmitted along a single traction force axis, but instead to be dissipated laterally. To quantify the additional impedance caused by this topological uncertainty, a dispersion transmission penalty weight was obtained based on the distribution characteristics of the basic traction transmission resistance parameters between each muscle region and adjacent interlayer muscle regions. The dispersion transmission penalty weight quantifies the mechanical transmission failure of diffuse and fragmented muscle tissue from the perspective of muscle bundle disintegration, preparing for subsequent correction of the basic traction transmission resistance parameters to accurately assess the traction transmission resistance.

[0039] In a preferred embodiment of the present invention, considering that the basic traction transmission resistance parameter is inversely proportional to the connection probability of muscle regions (the smaller the resistance, the greater the connection probability), to facilitate subsequent analysis of mechanical conduction dissipation, the interlayer muscle connection probability is first determined by negative correlation normalization. Furthermore, considering that information entropy can help quantify data distribution characteristics, taking the target region as an example, the more uniform and consistent the basic traction transmission resistance parameter is between the target region and different muscle regions in the next adjacent anatomical layer (i.e., the higher the entropy value), the more likely there are multiple dissipation directions in the traction force flow in the target region, indicating mechanical conduction dissipation. Conversely, the lower the entropy value, the more likely there is a clear main conduction path for the traction force flow, indicating concentrated mechanical conduction. Based on this, the method for obtaining the distributed transmission penalty weight includes: Between different muscle regions, the negative correlation normalization result of the basic traction transmission resistance parameter is used as the inter-layer muscle connection probability; for each muscle region, the normalization result of the entropy of the inter-layer muscle connection probability between it and each muscle region in the airway CT image of the adjacent anatomical layer is added with a constant 1 and used as the dispersion transmission penalty weight.

[0040] As an example, taking the target region as an example, the basic traction transfer resistance parameter between the target region and each muscle region in the next adjacent anatomical layer is mapped as the independent variable x to the exponential function exp(-x) with the natural constant e as the base. The mapping result is further divided by the sum of all mapping results to obtain the interlayer muscle connection probability between the target region and each muscle region in the next adjacent anatomical layer; the sum of the interlayer muscle connection probabilities between the target region and all muscle regions in the next adjacent anatomical layer is 1. Further calculate the entropy of the interlayer muscle connectivity probability between the target region and all muscle regions in the next adjacent anatomical layer. Normalize the entropy by dividing it by the maximum entropy value corresponding to all muscle regions in the anatomical layer where the target region is located. Add a constant 1 to the normalized entropy result to obtain the dispersion transmission penalty weight of the target region. In other examples, the implementer can also use a preset dissipation sensitivity weight, such as 0.5, to weight the normalized entropy result, and then add a constant 1 to obtain the dispersion transmission penalty weight of the target region.

[0041] Step S203: The basic traction transmission resistance parameter is weighted using the distributed transmission penalty weight to obtain the traction transmission resistance coefficient between each muscle region and each muscle region in the airway CT image of the adjacent anatomical layer.

[0042] The basic traction transfer resistance parameter between the target region and each muscle region in the next adjacent anatomical layer is further multiplied by the dispersion transfer penalty weight of the target region to obtain the traction transfer resistance parameter between the target region and each muscle region in the next adjacent anatomical layer.

[0043] By changing the target area, the traction transmission resistance coefficient between each muscle region and the airway CT image of the next adjacent anatomical layer can be obtained.

[0044] Since force transmission within biological tissues tends to follow the path of least resistance, in order to simulate the most efficient transmission mode of muscle bundles under extreme stress, it is necessary to find the globally optimal transmission pathway in the force transmission relationship between many-to-many interlayer muscle regions. Therefore, this invention, after obtaining the traction transmission resistance coefficient between different muscle regions, further performs interlayer matching on different muscle regions based on the traction transmission resistance coefficient to obtain the hierarchical transmission loss coefficient of each anatomical layer. The hierarchical transmission loss coefficient quantifies the force transmission impedance of a single anatomical surface, preparing for subsequent analysis of axial transmission loss along the preset traction force axis to assess airway risk characteristics.

[0045] Preferably, in one embodiment of the present invention, considering that bipartite graph matching forces the establishment of a one-to-one mechanical transmission chain, and that breakage can be handled through virtual nodes, the problem of connection competition due to mismatch in the number of muscle regions in upper and lower anatomical layers can be effectively solved, preventing repeated calculation of the load-bearing capacity of the same muscle region; and the tensile transmission resistance coefficient can help evaluate path transmission loss; therefore, interlayer matching of different muscle regions based on the tensile transmission resistance coefficient includes: Between adjacent anatomical layers, based on the bipartite graph matching algorithm and the traction transmission resistance coefficient between different muscle regions, all matching pairs between all different muscle regions are obtained.

[0046] In a preferred embodiment of the present invention, the bipartite graph matching algorithm includes at least the Hungarian algorithm.

[0047] As an example, between anatomical layer K and anatomical layer K+1, the muscle regions in anatomical layer K are considered as upper-level nodes (the total number of nodes is...). The muscle regions in anatomical layer K+1 are considered as lower-level nodes (the total number of nodes is...). When the number of nodes in the upper and lower layers is inconsistent, take... and The maximum value in is used as the matrix size L to construct an L×L square matrix. The matrix elements in the square matrix are set to P (which must be greater than the theoretically calculated maximum tensile resistance coefficient, with a value range of 2-10. In this example, 2 is used to represent extremely high resistance). Then, the traction transmission resistance coefficients between different muscle regions are mapped to the corresponding matrix positions for replacement. The first matrix element in the upper left corner represents the traction transmission resistance coefficient between muscle region number 1 in anatomical layer K and muscle region number 1 in anatomical layer K+1. The unreplaced matrix elements represent fracture zones or empty zones (corresponding matrix element P). Fracture zones indicate that the upper-layer node cannot find a lower-layer receiving node, and the muscle bundle is broken or disappears. Empty zones indicate that the lower-layer node has no corresponding upper-layer source and there is no mechanical activation. The Hungarian algorithm is used to solve the square matrix to obtain all matching pairs (matching pairs between upper and lower nodes) between different muscle regions. In other examples, the Kuhn-Munkres Algorithm can also be used. Both the Kuhn-Munkres Algorithm and the Hungarian algorithm are well-known techniques and will not be elaborated further.

[0048] It should be noted that among the determined matching pairs, matching pairs corresponding to fracture areas and empty areas need to be screened out. That is, the matching pairs must be interlayer matching results of muscle regions to prepare for the subsequent calculation of the layer transfer loss coefficient.

[0049] Further, based on the interlayer matching results of the muscle region, the hierarchical transfer loss coefficient of each anatomical layer is obtained.

[0050] Preferably, in one embodiment of the present invention, considering that simply summing the traction transfer resistance coefficients between matching muscle regions in anatomical layers may be affected by muscle mass—for example, a large, robust muscle bundle (with a large muscle region area) bears the main anti-collapse tension, while a small fragment (with a small muscle region area) contributes negligibly to the overall mechanics—if a simple arithmetic average is taken, the impedance value of the small fragment will greatly skew the overall result; therefore, the method for obtaining the layer transfer loss coefficient includes: In the airway CT images of each anatomical layer, the normalized result of the area of ​​each muscle region is used as the transfer weight. The transfer weight is used to weight the traction transfer resistance coefficient between the matching pairs to which the muscle region belongs, and the weighted sum is used as the layer transfer loss coefficient of the corresponding anatomical layer.

[0051] As an example, firstly, in the airway CT image of anatomical layer K, the area of ​​the region is represented by the number of pixels within the muscle region. Then, the area of ​​the region is normalized by dividing the sum of the areas of all muscle regions (or the area of ​​the slice sampling range) to obtain the transfer weight of each muscle region. Taking muscle region i in anatomical layer K as an example, the transfer weight is multiplied by the traction transfer resistance coefficient between the matching pairs to which muscle region i belongs. The corresponding multiplications of all muscle regions are accumulated and summed to obtain the layer transfer loss coefficient of anatomical layer K.

[0052] Step S3: Extract the airway risk characteristic parameters of the patient before anesthesia based on the variation characteristics of the hierarchical transmission loss coefficient of the anatomical layer along the preset traction force axis.

[0053] To more intuitively assess the relationship between anatomical location and the risk of airway obstruction due to airway collapse, this embodiment of the invention further extracts airway risk characteristic parameters of patients before anesthesia based on the variation characteristics of the hierarchical transmission loss coefficient of the anatomical layer along the preset traction force axis.

[0054] Preferably, in one embodiment of the present invention, an axial transmission loss curve is first fitted to transform discrete hierarchical data into a continuous anatomical-functional risk change curve, preparing for subsequent assessment of biological structural fracture and airway obstruction risk. Since the essence of airway risk characteristic assessment lies in finding the weakest link in the ability to resist collapse and assessing the overall structural characteristics of soft tissue, the maximum value of the curve can be used to characterize the maximum fracture risk, and the axial curve integral can be used to characterize the cumulative obstruction. Therefore, the method for obtaining airway risk characteristic parameters includes: The axial transmission loss curve is fitted based on the hierarchical transmission loss coefficient of each anatomical layer; the maximum value in the axial transmission loss curve is taken as the maximum airway rupture risk value, and the integral of the axial transmission loss curve is taken as the cumulative airway looseness value; the maximum airway rupture risk value and the cumulative airway looseness value are taken as airway risk characteristic parameters.

[0055] In a preferred embodiment of the present invention, considering that during anatomical slice sampling, due to CT volume effects or artifacts, the muscle boundary of a certain single-layer slice may exhibit slight jitter, leading to potential deviations in the muscle region of that layer. This, in turn, results in non-pathological spikes (noise) in the anatomical layer's layer transfer loss coefficient. Furthermore, the genioglossus muscle is a continuous entity with gradually changing mechanical properties, making drastic, abrupt changes between anatomical layers impossible. Therefore, after sorting the layer transfer loss coefficient along a preset traction force axis, a certain smoothing process is required to filter out noise and conform to the inherent properties of the tissue. Thus, the method for obtaining the axial transfer loss curve includes: The hierarchical transmission loss coefficients of each anatomical layer are sorted in order along the preset tensile force axis and then smoothed to obtain the axial transmission loss curve.

[0056] As an example, the width of the moving window is set to 3mm; the moving window is determined with any anatomical layer as the center, and each moving window contains 3 anatomical layers. The average of the hierarchical transfer loss coefficients of the 3 anatomical layers is used as the smoothed result of the hierarchical transfer loss coefficient of the anatomical layer corresponding to the center of the moving window; the cubic spline interpolation algorithm is used to fit all the smoothed results to obtain the axial transfer loss curve; the hierarchical transfer loss coefficients of the first and last anatomical layers are not smoothed by the moving window. Furthermore, the maximum value in the axial transmission loss curve is taken as the maximum airway rupture risk value R, and the integral of the axial transmission loss curve is taken as the cumulative airway looseness value L; R and L are the final extracted airway risk characteristic parameters; among them, the maximum airway rupture risk value R can help assess the anatomical layer where the force transmission chain is most likely to be interrupted after muscle relaxation, and the cumulative airway looseness value L helps quantify the overall degree of slippage of muscle bundle tissue. Both help clinicians determine whether external operations such as chin tuck can effectively transmit the lifting force to the base of the tongue through soft tissue, thereby formulating targeted airway management strategies before anesthesia induction.

[0057] It should be noted that moving smoothing, cubic spline interpolation algorithm for curve fitting, and the acquisition of maximum value and integral are all well-known techniques and will not be elaborated further.

[0058] It should be noted that airway risk characteristic parameters are only objective data parameters reflecting the physical conduction characteristics of airway soft tissue, used to assist doctors in understanding the anatomical structure of the patient's airway. The final clinical decision is made by the doctor in combination with other physical signs.

[0059] In summary, this invention first acquires CT images of the airway at each anatomical layer along a pre-defined traction force axis before anesthesia, and acquires all muscle regions in each airway CT image. The pre-defined traction force axis is the line connecting the mental spine of the mandible to the posterior wall of the airway. Between airway CT images of adjacent anatomical layers, based on the spatial misalignment and projection overlap between different muscle regions, the traction transmission resistance coefficient between different muscle regions is obtained. Based on the traction transmission resistance coefficient, inter-layer matching is performed on different muscle regions to obtain the hierarchical transmission loss coefficient for each anatomical layer. Based on the variation characteristics of the hierarchical transmission loss coefficient of the anatomical layers along the pre-defined traction force axis, airway risk characteristic parameters of the patient before anesthesia are extracted. This invention utilizes the spatial misalignment and overlap characteristics of inter-layer muscle regions to analyze muscle bundle distribution and force transmission impedance, and simulates stress transmission through inter-layer muscle region matching to analyze hierarchical transmission loss at anatomical layers, thereby extracting airway risk characteristics to assist in pre-anesthesia risk assessment.

[0060] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for extracting airway CT image features for pre-anesthesia risk assessment, characterized in that, The method includes: Acquire CT images of the airway at each anatomical layer along the pre-set traction force axis before anesthesia, and acquire all muscle regions in each airway CT image; wherein, the pre-set traction force axis is the direction of the line connecting the mental spine of the mandible to the posterior wall of the airway; Between airway CT images of adjacent anatomical layers, the traction transmission resistance coefficient between different muscle regions is obtained based on the spatial misalignment and projection overlap relationship between the different muscle regions. Based on the traction transmission resistance coefficient, inter-layer matching is performed on different muscle regions to obtain the layer transmission loss coefficient of each anatomical layer. Based on the variation characteristics of the hierarchical transmission loss coefficient of the anatomical layer along the preset traction force axis, airway risk characteristic parameters of the patient before anesthesia are extracted.

2. The method for airway CT image feature extraction for pre-anesthesia risk assessment according to claim 1, characterized in that, The method for obtaining the muscle region includes: Threshold segmentation was performed on the airway CT images of each anatomical layer to obtain all muscle pixels, and region connectivity detection was performed on the muscle pixels to obtain all muscle regions.

3. The method for airway CT image feature extraction for pre-anesthesia risk assessment according to claim 1, characterized in that, The method for obtaining the traction resistance coefficient includes: Between adjacent anatomical layers, based on the spatial distance and grayscale change information between the geometric centers corresponding to different muscle regions, as well as the overlapping area of ​​connected regions, the basic traction transmission resistance parameters between different muscle regions are obtained. For each muscle region, a distributed transfer penalty weight is obtained based on the distribution characteristics of the basic traction transfer resistance parameters between the muscle region and the airway CT images of adjacent anatomical layers. The basic traction transmission resistance parameter is weighted using the dispersed transmission penalty weight to obtain the traction transmission resistance coefficient between each muscle region and the airway CT image of the adjacent anatomical layer.

4. The method for airway CT image feature extraction for pre-anesthesia risk assessment according to claim 3, characterized in that, The method for obtaining the basic tensile resistance parameters includes: Between different muscle regions, the normalized result of the Euclidean distance between the corresponding geometric centers is used as the first resistance parameter, the negative correlation normalized result of the pixel mean of the voxel of the path connecting the corresponding geometric centers is used as the second resistance parameter, and the negative correlation normalized result of the Jaccard similarity coefficient between the corresponding pixel sets is used as the third resistance parameter. The first resistance parameter, the second resistance parameter, and the third resistance parameter are weighted and fused to obtain the basic traction transmission resistance parameters between different muscle regions.

5. The method for airway CT image feature extraction for pre-anesthesia risk assessment according to claim 3, characterized in that, The method for obtaining the distributed propagation penalty weight includes: Between different muscle regions, the negative correlation normalization result of the basic traction transmission resistance parameter is used as the interlayer muscle connection probability; for each muscle region, the normalization result of the entropy of the interlayer muscle connection probability between it and each muscle region in the airway CT image of the adjacent anatomical layer is added with a constant 1 and used as the dispersion transmission penalty weight.

6. The method for airway CT image feature extraction for pre-anesthesia risk assessment according to claim 1, characterized in that, Interlayer matching of different muscle regions based on the aforementioned traction transmission resistance coefficient includes: Between adjacent anatomical layers, based on the bipartite graph matching algorithm and the traction transmission resistance coefficient between different muscle regions, all matching pairs between all different muscle regions are obtained.

7. The method for airway CT image feature extraction for pre-anesthesia risk assessment according to claim 6, characterized in that, Bipartite graph matching algorithms include at least the Hungarian algorithm.

8. A method for extracting airway CT image features for pre-anesthesia risk assessment according to claim 6, characterized in that, The method for obtaining the hierarchical transmission loss coefficient includes: In the airway CT image of each anatomical layer, the normalized result of the area of ​​each muscle region is used as the transfer weight. The transfer weight is used to weight the traction transfer resistance coefficient between the matching pairs to which the muscle region belongs. The weighted sum is used as the layer transfer loss coefficient of the corresponding anatomical layer.

9. A method for extracting airway CT image features for pre-anesthesia risk assessment according to claim 1, characterized in that, The methods for obtaining the airway risk characteristic parameters include: An axial transmission loss curve is fitted based on the hierarchical transmission loss coefficient of each anatomical layer; the maximum value in the axial transmission loss curve is taken as the maximum airway rupture risk value, and the integral of the axial transmission loss curve is taken as the cumulative airway looseness value; the maximum airway rupture risk value and the cumulative airway looseness value are taken as airway risk characteristic parameters.

10. A method for extracting airway CT image features for pre-anesthesia risk assessment according to claim 9, characterized in that, The method for obtaining the axial transmission loss curve includes: The hierarchical transmission loss coefficients of each anatomical layer are sorted in order along the preset tensile force axis and then smoothed to obtain the axial transmission loss curve.

Images