Non-contact respiratory mechanics inverse reconstruction system based on 4d holographic field perception
By constructing a 4D point cloud sequence of the chest and abdominal surface through 4D holographic field perception, generating respiratory driving features and evaluating observation confidence, the stability and reliability issues of non-contact respiratory monitoring in existing technologies are solved, and stable output of airway resistance and pleural cavity pressure changes is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUNAN XIAXIA MEDICAL EQUIPMENT CO LTD
- Filing Date
- 2026-04-09
- Publication Date
- 2026-06-23
AI Technical Summary
Existing non-contact respiratory monitoring methods are unable to fully express process information such as the inspiratory-exhalation transition and expansion rate, and lack quantification and explicit constraints on the reliability of observations, resulting in unstable mechanical parameter outputs.
Based on 4D holographic field perception, a 4D point cloud sequence of time-varying morphology of the chest and abdomen surface is constructed. Respiratory drive features are generated by geodesic length, curvature tensor evolution and area expansion rate. Viscoelastic dynamics are inversely reconstructed by combining observation confidence and outputting changes in airway resistance and pleural cavity pressure.
It improves the completeness of the expression of inhalation-exhalation transition and expansion rate, enhances the stability and reliability of mechanical parameter output, and reduces the disturbance of the results by observation fluctuations.
Smart Images
Figure CN121997843B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of holographic respiratory analysis technology, and more specifically, to a non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception. Background Technology
[0002] Respiratory mechanics parameters (such as airway resistance, compliance, and changes in intrapleural pressure) are crucial for assessing ventilation status, identifying the risk of respiratory depression, and guiding perioperative respiratory management. In recent years, non-contact sensing methods such as millimeter-wave radar have enabled the acquisition of motion information of the chest and abdominal surfaces over time without the need for wearing sensors, providing a new technological approach for continuous respiratory assessment in clinical monitoring, postoperative recovery, and home monitoring scenarios.
[0003] Existing non-contact respiratory monitoring and respiratory parameter inference methods mostly use single-point displacement, local amplitude, or single time-series curves as observation inputs. When the deformation of the chest and abdominal surface exhibits local fluctuations, uneven expansion, or different regional response differences, it is difficult to form a more complete boundary description of process information such as inspiratory-expiratory transition and expansion rate, thus limiting the conditional expression for further respiratory mechanics inverse reconstruction. Furthermore, in continuous monitoring, the observation stability and effective information content may vary across different respiratory cycles. Without quantifying the reliability of observations and explicit constraints during the inversion process, the mechanical parameter outputs are prone to instability or difficulty in assessing reliability under the influence of fluctuating segments. Therefore, it is necessary to develop a non-contact respiratory mechanics inverse reconstruction method that can extract structured driving features from the time-varying morphology of the chest and abdominal surface and combine them with observation reliability constraints. Summary of the Invention
[0004] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception, comprising:
[0005] The 4D holographic sensing and breathing observation module receives the echo sequence of the chest and abdomen region from the millimeter-wave sparse aperture array, generates a 4D point cloud sequence of the time-varying morphology of the chest and abdomen surface frame by frame, and constructs a breathing observation sequence of the chest and abdomen surface; the breathing observation sequence includes a geodesic length sequence, an area dilatation rate sequence, and a velocity field sequence of the chest and abdomen region.
[0006] The geodesic and curvature energy correlation module generates respiratory drive features by analyzing the correlation between the geodesic length variation attributes and the curvature tensor evolution sequence energy variation attributes.
[0007] The respiratory drive confidence assessment module obtains the observation confidence based on the temporal changes of the chest and abdominal regions in the respiratory drive characteristics;
[0008] The viscoelastic dynamics inverse reconstruction module inputs the respiratory drive characteristics into an inverse reconstruction network with nonlinear viscoelastic respiratory dynamics constraints, and outputs a set of mechanical parameters of airway resistance and pleural cavity pressure changes.
[0009] Preferably, the method for constructing respiratory observation sequences on the thoracic and abdominal surfaces includes:
[0010] By analyzing the adjacency relationships between different points in each sampling frame, a triangular mesh set of the thoracic and abdominal surface is constructed.
[0011] Calculate the local principal curvature at each point and converge them to generate a curvature tensor evolution sequence;
[0012] Construct a set of monitoring anchor point trajectories for the surface regions of the thorax and abdomen, and calculate the geodesic length sequence, area dilation rate sequence, and velocity field sequence of the thorax and abdomen regions.
[0013] By combining the geodesic length sequence, the area dilation rate sequence, and the velocity field sequence of the thoracic and abdominal regions, a respiratory observation sequence is generated.
[0014] Preferably, the method for generating the curvature tensor evolution sequence includes:
[0015] In each frame, the point set is traversed, and the points obtained during the traversal are marked as target points;
[0016] Calculate the normal vector of the target point, construct a normal plane coordinate system perpendicular to the normal vector, and construct the edge vector between the target point and each point in the list of neighboring points, denoted as the neighboring edge vector;
[0017] Calculate the difference between the normal vectors of the neighboring points and the target point, and project the neighboring edge vectors and the difference between the normal vectors onto the normal plane coordinate system to obtain the projection vectors of the neighboring edges and the projection vectors of the difference between the normal vectors.
[0018] Calculate the outer product matrix of the projection vectors of different neighboring edges separately and sum them to obtain the projection matrix of the neighboring edges;
[0019] Calculate the product matrix of the projection vector of the neighboring edge formed by the same neighboring point and the normal difference projection vector, and sum the product matrices of different neighboring points to obtain the normal response matrix;
[0020] The curvature tensor matrix of the target point is calculated based on the projection matrix of the adjacent edge and the normal response matrix. The curvature tensor matrix is then symmetricized and decomposed into eigenvalues to obtain two eigenvalues as the local principal curvatures of the target point.
[0021] The local principal curvature of each point in the combined point set under different adjacent frames is used to obtain the curvature tensor. The curvature tensors are then spliced together in the order of the frame numbers to obtain the curvature tensor evolution sequence.
[0022] Preferably, the method for generating respiratory observation sequences includes:
[0023] In each frame's point set, select the triangular meshes located in the thoracic region and the abdominal region respectively, filter the triangular meshes located in the three target regions respectively, and mark them as target meshes;
[0024] Mark the vertices of different target grids within the target area as anchor points of the target area;
[0025] For each frame, enumerate the anchor point pairs located in the thoracic region and the abdominal region respectively. For each anchor point pair in the region, calculate all the side lengths of the triangular mesh on the thoracic and abdominal surfaces of each frame. Randomly select one anchor point between the anchor point pairs and start searching for another anchor point along the edge to obtain the search path.
[0026] Enumerate the search paths to obtain the anchor point pairs; accumulate the lengths of all sides in the anchor point pairs to obtain the path length.
[0027] For the same pair of anchor points, the path of the anchor point pair with the minimum path length is selected as the geodesic. During the generation of the geodesic, the order of the points passed through is combined to obtain the geodesic path.
[0028] For each pair of anchor points, the geodesic path is solved repeatedly on consecutive frames, and the path length is accumulated to obtain the geodesic length sequence.
[0029] Preferably, the method for calculating the area expansion rate sequence includes:
[0030] Using different geodesics on the surface of the chest and abdomen of each frame as the center, geodesic strips are obtained by expanding to both sides with a fixed width.
[0031] The area of the geodesic strip region is statistically analyzed. For the same anchor point pair in different frames, the difference of the geodesic strip region in adjacent frames is calculated.
[0032] The area expansion rate is calculated by comparing the difference result with the area of the geodesic strip region in the previous frame in the adjacent frame. The area expansion rates are then combined according to the development order of the frame number to obtain the area expansion rate sequence.
[0033] Preferably, the method for calculating the velocity field sequence in the chest and abdomen region includes:
[0034] Calculate the displacement vector of each point in adjacent frames, and then calculate the dot product of the unit normal vector and the displacement vector, which is denoted as the normal displacement.
[0035] The triangular meshes constructed by each point within the set of integration points are recorded as the list of participating meshes for the corresponding points.
[0036] Calculate the cumulative face value of all triangular meshes in the mesh list, divide the cumulative face value into three equal parts and normalize it to obtain the area weight of the corresponding point;
[0037] The normal displacements of different points in the surface regions of the thorax and abdomen are fused by area weighting, and the frame interval between the weighted fusion result and the adjacent frames is calculated to obtain the displacement velocity of the thorax and the displacement velocity of the abdomen.
[0038] By combining the thoracic displacement velocity and the abdominal displacement velocity in the order of development according to the frame number, the velocity field of the thoracic and abdominal regions is obtained.
[0039] Preferably, the method for generating respiratory drive features includes:
[0040] Identify the monotonic intervals of the geodesic length sequence, and include the monotonic intervals of increasing geodesic length and decreasing monotonic intervals into the candidate sets for the inspiratory and expiratory segments, respectively.
[0041] The rising and falling slopes of the curvature tensor evolution sequence energy are calculated in the candidate sets of the inspiratory and expiratory segments, respectively, to generate the expansion rate indicator sequence.
[0042] By analyzing the consistency between the slope change properties in the expansion rate indicator sequence and the growth in the area expansion rate sequence, a set of key moments including the initiation of inhalation, the pause, and the initiation of exhalation is constructed.
[0043] Extract the duration of inhalation and exhalation from the critical moment set;
[0044] Extract the maximum displacement velocity of the thoracic and abdominal regions in each respiratory cycle from the velocity field of the thoracic and abdominal regions to obtain the peak expansion rate of the thoracic and abdominal regions.
[0045] Within each respiratory cycle, the area expansion rate sequence was statistically analyzed, and the area expansion of the thoracic and abdominal regions was obtained and accumulated. The accumulated result was recorded as the body surface expansion proxy amount.
[0046] By integrating inspiratory duration, expiratory duration, peak expansion rate, inspiratory-expiratory transition steepness, and surface expansion proxy, respiratory drive characteristics of the thoracic and abdominal surface regions were obtained.
[0047] Preferably, the method for generating the expansion rate indication sequence includes:
[0048] For the thoracic and abdominal regions, the curvature energy is obtained by squaring and summing the local principal curvature of each point in the curvature tensor evolution sequence.
[0049] The curvature energy of all points within the same region is combined according to the frame number;
[0050] The curvature energy sequence is divided into interval slices according to the candidate sets of inspiratory and expiratory segments to obtain the energy subsequence sets of inspiratory and expiratory segments respectively.
[0051] For each inspiratory segment energy subsequence and expiratory segment energy subsequence, sliding fitting is performed within the segment to obtain the rising slope sequence and the falling slope sequence, respectively.
[0052] The ascending slope sequence and the descending slope sequence are spliced together, and the splicing result is recorded as the expansion rate indicator sequence.
[0053] Preferably, the method for constructing a set of key moments including the initiation of inhalation, the pause, and the initiation of exhalation includes:
[0054] Identify consecutive non-negative segments and consecutive non-positive segments in the area expansion rate sequence, and match the timestamps corresponding to the consecutive non-negative segments with the timestamps of the rising slope.
[0055] Mark the earliest time obtained from the matching as the inhalation start time, and the latest time obtained from the matching as the inhalation peak time;
[0056] Match the timestamps corresponding to consecutive non-positive segments with the timestamps of the descending slope, and mark the earliest matching time as the exhalation start time;
[0057] By integrating the key moments of inhalation onset, peak inhalation, and exhalation onset, a set of key moments is obtained.
[0058] Preferably, the method for settling the mechanical parameters of output airway resistance and pleural cavity pressure changes includes:
[0059] The changes in intrapleural pressure and airway resistance are defined as state variables of the network output, and respiratory drive characteristics are used as boundary excitation observations.
[0060] The breathing drive features and observation confidence are used as boundary excitation inputs to inversely reconstruct the network, and the observation confidence is set as the sample weight of the training loss to generate a weighted joint loss.
[0061] By analyzing the correlation between the descent rate of the weighted joint loss during the iteration process and the smoothness of the state variable time series, the network convergence is determined and the state variables are output, thus obtaining the set of mechanical parameters.
[0062] The technical effects and advantages of this invention: a non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception.
[0063] (1) This invention constructs a 4D point cloud and triangular mesh observation sequence of the chest and abdomen, and generates respiratory driving features based on the correlation analysis of geodesic length change and curvature energy evolution. At the same time, it introduces area expansion and velocity field statistics to form a body surface expansion proxy quantity. The advantage is that it transforms morphological observation into structured boundary excitation, improves the completeness of the expression of inhalation and exhalation conversion and expansion rate, and enhances the usability and interpretability of subsequent mechanical parameter inversion.
[0064] (2) The present invention generates observation confidence based on the time change of respiratory drive characteristics and maps it to the sample weights of training loss to form a weighted joint loss. At the same time, the convergence output is determined by the correlation criterion between loss decrease and smoothness of state variables. The advantage is that the observation reliability is explicitly incorporated into the inversion constraint, which improves the stability and reliability of outputs such as airway resistance and pleural pressure changes, and reduces the disturbance propagation of observation fluctuations on the results. Attached Figure Description
[0065] Figure 1 This is a schematic diagram of the system structure of the non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception of the present invention.
[0066] Figure 2 This is a schematic diagram of the method flow of the non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception of the present invention.
[0067] Figure 3 This is a schematic diagram of the method for generating respiratory drive features in the non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to the present invention. Detailed Implementation
[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0069] This application provides a non-contact respiratory mechanics inverse reconstruction system based on 4D holographic field perception. A 4D point cloud of the chest and abdomen is constructed using a millimeter-wave sparse array, along with respiratory observation sequences. Respiratory drive features are generated based on the correlation between geodesic length and curvature energy, and the observation confidence level is evaluated. These drive features and confidence levels are then introduced into an inverse network with viscoelastic dynamic constraints. Weighted physical residuals are used to train and invert the mechanical parameters of airway resistance and pleural pressure changes. This invention enhances the completeness of the respiratory drive expression and improves the stability and reliability of the output mechanical parameters by transforming 4D morphological observations into structured boundary excitations and using observation confidence levels to weight and constrain the inversion process.
[0070] Please see Figure 1 , Figure 2 and Figure 3 In this embodiment of the invention, the non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception is implemented in detail through the following modules:
[0071] The 4D holographic sensing and respiratory observation module receives the echo sequence of the chest and abdomen region from the millimeter-wave sparse aperture array, generates a 4D point cloud sequence of the time-varying morphology of the chest and abdomen surface frame by frame, and constructs a respiratory observation sequence of the chest and abdomen surface; the respiratory observation sequence includes geodesic length sequence, area dilation rate sequence and chest and abdomen region velocity field sequence.
[0072] Methods for constructing respiratory observation sequences on the thoracic and abdominal surfaces include:
[0073] By analyzing the adjacency relationships between different points in each sampling frame, a triangular mesh set of the thoracic and abdominal surface is constructed.
[0074] The specific methods for constructing a triangular mesh set for the thoracic and abdominal surfaces include:
[0075] The coordinates of three-dimensional points in a set of points are quantized into discrete integer coordinates, and the discrete integer coordinates are combined into the spatial key of the point.
[0076] Iterate through all spatial keys in the point record set and construct a point identifier list for each spatial key; the point identifier list uses a dynamic array or linked list structure to support the aggregation of multiple points within the same spatial key;
[0077] An example of constructing a list of point identifiers for spatial keys is illustrated below:
[0078] A set of point records for a certain frame includes point P1 with coordinates (x, y, z) = (0.041, 0.018, 0.306), point P2 with coordinates (0.043, 0.017, 0.312), point P3 with coordinates (0.042, 0.019, 0.318), point P4 with coordinates (0.041, 0.019, 0.325), and point P5 with coordinates (0.098, 0.016, 0.303). Discretize the coordinates as follows: Given the space keys, we get: P1: (2,0,15), P2: (2,0,15), P3: (2,0,15), P4: (2,0,16), P5: (4,0,15); then the point identifier list for space key (2,0,15) is: [P1, P2, P3]; the point identifier list for space key (2,0,16) is: [P4]; the point identifier list for space key (4,0,15) is: [P5]; where P is the point identifier.
[0079] Each three-dimensional grid after discretization of three-dimensional coordinates is denoted as a grid cell. For each grid cell containing a spatial key, 26 adjacent grid cells in three-dimensional space are combined to obtain the offset set of the spatial key. Other spatial keys in the offset set of each spatial key are retrieved, and an enumeration table of adjacent spatial keys of the corresponding spatial key is constructed.
[0080] Calculate the list of in-key points for each point within the same spatial key, and the list of out-of-key points for all points in the enumeration table of adjacent spatial keys of the spatial key. Integrate the in-key point list and the out-of-key point list to obtain the candidate point list. Calculate the distance between the target point and each point in the candidate point list, and retain the 4 closest points.
[0081] Traverse each point, marking the traversed points as the target point. Enumerate adjacent point pairs in the list of neighboring points of the target point, and combine each adjacent point pair with the target point in turn to construct a candidate triangular mesh for the target point and integrate them to obtain the mesh set of the target point. For example, the undirected adjacency list of the target point P0 includes P1, P2, and P3. Then, enumerating the adjacent point pairs yields (P1, P2), (P1, P3), and (P2, P3), resulting in the candidate triangular meshes: T12=(P0, P1, P2), T13=(P0, P1, P3), and T23=(P0, P2, P3). Here, T represents the triangular mesh identifier, and the number after the identifier identifies the number of the adjacent point pair of the triangular mesh.
[0082] Traverse the set of points and mark the points obtained as target points. Construct a reference triangular mesh based on the two nearest points of the target point. Calculate the normal vector of the reference triangular mesh and each candidate triangular mesh to obtain the reference vector and candidate vector. Eliminate the candidate triangular mesh corresponding to the candidate vector whose direction is inconsistent with the reference vector.
[0083] In this case, the vector directions are inconsistent. The dot product of the reference vector and the candidate vector is calculated. If the value of the dot product is greater than 0, it means that the vector directions are consistent; otherwise, the vector directions are inconsistent.
[0084] To illustrate the calculation of the normal vector, consider the target point P0 and its nearest neighbors P1 and P2, whose coordinates are: P0=(0,0,1000), P1=(20,0,1001), and P2=(0,20,1002). The triangular mesh is (P0,P1,P2). The edge vectors v1 and v2 are calculated as: v1=P1−P0=(20,0,1) and v2=P2−P0=(0,20,2). The normal vector n... In calculating the edge vectors, the method of calculating the difference between the coordinates of non-target points and target points is adopted.
[0085] Integrate the candidate mesh sets for each point and remove duplicates to obtain a triangular mesh set;
[0086] Calculate the local principal curvature at each point and converge them to generate a curvature tensor evolution sequence;
[0087] In each frame, the point set is traversed, and the points obtained during the traversal are marked as target points;
[0088] Calculate the normal vector of the target point, construct a normal plane coordinate system perpendicular to the normal vector, and construct the edge vector between the target point and each point in the list of neighboring points, denoted as the neighboring edge vector;
[0089] The normal plane coordinate system is first constructed by building a normal plane including the corresponding points based on the normal vectors. A two-dimensional coordinate system is then constructed in the normal plane with the corresponding points as the origin, which is denoted as the normal plane coordinate system.
[0090] Calculate the difference between the normal vectors of the neighboring points and the target point, and project the neighboring edge vectors and the difference between the normal vectors onto the normal plane coordinate system to obtain the projection vectors of the neighboring edges and the projection vectors of the difference between the normal vectors.
[0091] Calculate the outer product matrix of the projection vectors of different neighboring edges separately and sum them to obtain the projection matrix of the neighboring edges;
[0092] For example, the projection vectors of the neighboring edges of the target point are respectively , , , ;
[0093] The outer product matrices are respectively , , , The cumulative projection matrix of the nearest edge is denoted as . ;
[0094] Calculate the product matrix of the projection vector of the neighboring edge formed by the same neighboring point and the normal difference projection vector, and sum the product matrices of different neighboring points to obtain the normal response matrix;
[0095] For example, the target point has four normal difference projection vectors, namely... , , , ;
[0096] The product matrices are respectively , , , The summation yields the normal response matrix, denoted as... ;
[0097] The curvature tensor matrix of the target point is calculated based on the projection matrix of the adjacent edge and the normal response matrix. The curvature tensor matrix is then symmetricized and decomposed into eigenvalues to obtain two eigenvalues as the local principal curvatures of the target point.
[0098] The following are examples illustrating the local principal curvature and principal direction:
[0099] Through formula The curvature tensor matrix K of the target point is calculated as follows:
[0100] ;
[0101] Symmetric simplification yields the simplified curvature tensor matrix. ;
[0102] For the simplified curvature tensor matrix Perform eigenvalue decomposition to obtain two eigenvalues. , Then the two eigenvalues are denoted as the local principal curvatures. and According to the local principal curvature and The reverse direction angle is obtained , , taking them as the two main directions;
[0103] The local principal curvature of each point in the set of combined points under different adjacent frames is used to obtain the curvature tensor. The curvature tensors are then spliced together in the order of the frame numbers to obtain the curvature tensor evolution sequence.
[0104] Construct a set of monitoring anchor point trajectories for the surface regions of the thorax and abdomen, and calculate the geodesic length sequence, area dilation rate sequence, and velocity field sequence of the thorax and abdomen regions.
[0105] Methods for generating respiratory observation sequences include:
[0106] In each frame's point set, select the triangular meshes located in the thoracic region and the abdominal region respectively, filter the triangular meshes located in the three target regions respectively, and mark them as target meshes;
[0107] Mark the vertices of different target grids within the target area as anchor points of the target area;
[0108] For each frame, enumerate the anchor point pairs located in the thoracic region and the abdominal region respectively. For each anchor point pair in the region, calculate all the side lengths of the triangular mesh on the thoracic and abdominal surfaces of each frame. Randomly select one anchor point between the anchor point pairs and start searching for another anchor point along the edge to obtain the search path.
[0109] Enumerate the search paths to obtain the anchor point pairs; accumulate the lengths of all sides in the anchor point pairs to obtain the path length.
[0110] For the same pair of anchor points, the path of the anchor point pair with the minimum path length is selected as the geodesic. During the generation of the geodesic, the order of the points passed through is combined to obtain the geodesic path.
[0111] For each pair of anchor points, the geodesic path is repeatedly solved on consecutive frames, and the path length is accumulated to obtain the geodesic length sequence.
[0112] Methods for calculating the area expansion rate sequence include:
[0113] Using different geodesics on the surface of the chest and abdomen of each frame as the center, geodesic strips are obtained by expanding to both sides with a fixed width.
[0114] In this embodiment, the fixed width is taken as 4 times the average side length of all triangular meshes;
[0115] The area of the geodesic strip region is statistically analyzed. For the same anchor point pair in different frames, the difference of the geodesic strip region in adjacent frames is calculated.
[0116] The area expansion rate is calculated by comparing the difference result with the area of the geodesic strip region in the previous frame in the adjacent frame number. The area expansion rates are then combined according to the development order of the frame number to obtain the area expansion rate sequence.
[0117] Methods for calculating velocity field sequences in the chest and abdomen regions include:
[0118] Calculate the displacement vector of each point in adjacent frames, and then calculate the dot product of the unit normal vector and the displacement vector, which is denoted as the normal displacement.
[0119] The triangular meshes constructed by each point within the set of integration points are recorded as the list of participating meshes for the corresponding points.
[0120] Calculate the cumulative face value of all triangular meshes in the mesh list, divide the cumulative face value into three equal parts and normalize it to obtain the area weight of the corresponding point;
[0121] The normal displacements of different points in the surface regions of the thorax and abdomen are fused by area weighting, and the frame interval between the weighted fusion result and the adjacent frames is calculated to obtain the displacement velocity of the thorax and the displacement velocity of the abdomen.
[0122] By combining the thoracic displacement velocity and the abdominal displacement velocity in the order of development according to the frame number, the velocity field of the thoracic and abdominal regions is obtained.
[0123] By combining geodesic length sequences, area dilatation rate sequences, and velocity field sequences in the thoracic and abdominal regions, a respiratory observation sequence is generated.
[0124] The geodesic and curvature energy correlation module generates respiratory drive features by analyzing the correlation between the geodesic length variation attributes and the curvature tensor evolution sequence energy variation attributes.
[0125] Methods for generating respiratory drive features include:
[0126] Identify the monotonic intervals of the geodesic length sequence, and include the monotonic intervals of increasing geodesic length and decreasing monotonic intervals into the candidate sets for the inspiratory and expiratory segments, respectively.
[0127] The rising and falling slopes of the curvature tensor evolution sequence energy are calculated in the candidate sets of the inspiratory and expiratory segments, respectively, to generate the expansion rate indicator sequence.
[0128] Methods for generating expansion rate indicator sequences include:
[0129] For the thoracic and abdominal regions, the curvature energy is obtained by squaring and summing the local principal curvature of each point in the curvature tensor evolution sequence.
[0130] The curvature energy of all points within the same region is combined according to the frame number;
[0131] Among them, the effective area of the chest and abdomen is the outer polygon of the chest and abdomen enclosed by all anchor points;
[0132] The curvature energy sequence is divided into interval slices according to the candidate sets of inspiratory and expiratory segments to obtain the energy subsequence sets of inspiratory and expiratory segments respectively.
[0133] For each inspiratory segment energy subsequence and expiratory segment energy subsequence, sliding fitting is performed within the segment to obtain the rising slope sequence and the falling slope sequence, respectively.
[0134] The ascending slope sequence and the descending slope sequence are spliced together, and the splicing result is recorded as the expansion rate indicator sequence.
[0135] By analyzing the consistency between the slope change properties in the expansion rate indicator sequence and the growth in the area expansion rate sequence, a set of key moments including the initiation of inhalation, the pause, and the initiation of exhalation is constructed.
[0136] Methods for constructing a set of key moments including the initiation of inhalation, the pause, and the initiation of exhalation include:
[0137] Identify consecutive non-negative segments and consecutive non-positive segments in the area expansion rate sequence, and match the timestamps corresponding to the consecutive non-negative segments with the timestamps of the rising slope.
[0138] Mark the earliest time obtained from the matching as the inhalation start time, and the latest time obtained from the matching as the inhalation peak time;
[0139] Match the timestamps corresponding to consecutive non-positive segments with the timestamps of the descending slope, and mark the earliest matching time as the exhalation start time;
[0140] By integrating the key moments of inspiratory initiation, peak inspiratory activity, and expiratory initiation, a set of key moments is obtained.
[0141] Extract the duration of inhalation and exhalation from the critical moment set;
[0142] The duration of the inhalation from the start of inhalation to the first pause is recorded as the inhalation duration; the duration of the exhalation from the start of exhalation to the start of the next exhalation is recorded as the exhalation duration.
[0143] The respiratory cycle of a region is determined based on the interval between the inspiratory onset and expiratory peak times in the critical moment set.
[0144] Extract the maximum displacement velocity of the thoracic and abdominal regions in each respiratory cycle from the velocity field of the thoracic and abdominal regions to obtain the peak expansion rate of the thoracic and abdominal regions.
[0145] For each inspiratory peak moment of the expansion rate indicator sequence, the absolute values of the average rising slope and the average falling slope within the neighborhood window are accumulated, and the accumulated result is recorded as the inspiratory-expiratory transition steepness.
[0146] The neighborhood window length is divided into four equal parts based on the average respiratory cycle.
[0147] Within each respiratory cycle, the area expansion rate sequence was statistically analyzed, and the area expansion of the thoracic and abdominal regions was obtained and accumulated. The accumulated result was recorded as the body surface expansion proxy amount.
[0148] By integrating inspiratory duration, expiratory duration, peak expansion rate, inspiratory-expiratory transition steepness, and surface expansion proxy, respiratory drive characteristics of the thoracic and abdominal surface regions were obtained.
[0149] In existing technologies, non-contact respiratory monitoring often characterizes respiration using single-point displacement or a single curve, which makes it difficult to form stable boundary conditions suitable for mechanical inversion when faced with complex deformations of the chest and abdominal surfaces. This invention constructs a 4D point cloud and triangular mesh observation sequence for the chest and abdomen, and generates respiratory driving features based on the correlation analysis of geodesic length changes and curvature energy evolution. At the same time, it introduces area expansion and velocity field statistics to form surface expansion surrogate quantities. The advantage is that it transforms morphological observation into structured boundary excitation, improves the completeness of the expression of inhalation-exhalation conversion and expansion rate, and enhances the usability and interpretability of subsequent mechanical parameter inversion.
[0150] The respiratory drive confidence assessment module obtains the observed confidence based on the temporal changes of the chest and abdominal regions in respiratory drive characteristics;
[0151] Calculate the difference in inspiratory onset, peak expansion, inspiratory duration, and expiratory duration for the thoracic and abdominal surface regions in each respiratory cycle.
[0152] The different time differences and time intervals within the same respiratory cycle are weighted and fused, and the weighted fusion result is recorded as the mismatch magnitude; in this embodiment, the weights are all set to 0.25;
[0153] The difference between the mismatch magnitude and the preset mismatch magnitude threshold is calculated, and the ratio of the calculated result to the mismatch magnitude threshold is calculated. The calculated result is recorded as the observation confidence level.
[0154] In this embodiment, the mismatch threshold is taken as the mismatch magnitude value at the 20th percentile of the historical mismatch magnitude test data in ascending order.
[0155] The viscoelastic dynamics inverse reconstruction module inputs respiratory drive characteristics into an inverse reconstruction network with nonlinear viscoelastic respiratory dynamics constraints, and outputs a set of mechanical parameters of airway resistance and pleural cavity pressure changes.
[0156] Methods for settling mechanical parameters of airway resistance and changes in intrapleural pressure include:
[0157] The changes in intrapleural pressure and airway resistance are defined as state variables of the network output, and respiratory drive characteristics are used as boundary excitation observations.
[0158] Among them, the change in pleural cavity pressure represents a discrete sampling sequence of pleural pressure within a respiratory cycle, and airway resistance represents the impedance parameter of the relationship between airflow and pressure difference.
[0159] The breathing drive features and observation confidence are used as boundary excitation inputs to inversely reconstruct the network, and the observation confidence is set as the sample weight of the training loss to generate a weighted joint loss.
[0160] In this embodiment, the perioperative monitoring scenario is based on the patient lying supine after anesthesia induction in the operating room and being ventilated by a ventilator with volume-controlled ventilation (tidal volume of approximately 6–8 mL / kg and respiratory rate of 10–14 breaths per minute). A sparse aperture MIMO millimeter-wave radar is deployed at approximately 0.6–1.0 m in front of the patient's chest to acquire 4D point clouds of the chest and abdomen. Side respiratory driving features (peak expansion rate, inspiratory duration, expiratory duration, inspiratory-expiratory transition steepness, surface expansion surrogate quantity, and observation confidence) are extracted for each respiratory cycle and used as boundary excitation inputs for the network. The inverse reconstruction network adopts a simplified PINN structure that combines a time encoder and a physical constraint state head.
[0161] The time encoder uses a double-gated cyclic unit to encode the respiratory drive feature sequence formed by respiratory drive features of several consecutive frames. The encoding specifically includes: the number of hidden units is 64, and the sequence length is the frame segment corresponding to the most recent 2-3 respiratory cycles. The status head outputs the parameter trajectory of airway resistance and pleural cavity pressure changes on the same time axis, and embeds the discrete difference residual of nonlinear viscoelastic dual-chamber respiratory dynamics inside the network as a physical constraint. The discrete difference residual of nonlinear viscoelastic dual-chamber respiratory dynamics is obtained by discretizing the dynamic equation describing the pressure-volume-flow coupling relationship between alveolar cavity and pleural cavity into frame-by-frame recursive constraints according to the radar frame interval, and substituting the resistance compliance and pressure changes output by the network into the calculation of the frame-by-frame inconsistency deviation.
[0162] During model training, under non-invasive conditions, the tidal volume waveform and airway pressure waveform that can be directly read from the ventilator side are used as external reference signals, and combined with respiratory drive characteristics, observation fitting terms are generated.
[0163] By mapping observation confidence levels in segments, sample weights are generated. In this embodiment, when the observation confidence level is ≥0.8, the weight is 1.0; when the observation confidence level is in the range of 0.6–0.8, the weight is 0.6; and when the observation confidence level is <0.6, the weight is 0.2, forming a weighted joint loss.
[0164] The total loss for each training sample is defined as: the sample weight multiplied by the sum of the observation fit term and the physical residual term; the Adam optimizer is used for iterative training in mini-batch sampling by breathing cycle until the weighted joint loss converges stably.
[0165] By analyzing the correlation between the decreasing rate of the weighted joint loss during the iteration process and the smoothness of the state variable time series, network convergence and output of state variables are determined.
[0166] Specifically, when the weighted joint loss decreases by less than 5% in five consecutive iterations and continues to decrease, the rate of decrease is considered converged; when the difference between each state variable in the state variable vector sequence in adjacent frames decreases by less than 5% in five consecutive iterations and continues to decrease, the time series of state variables is considered smooth converged; when both the rate of decrease converges and the time series of smooth converges occur simultaneously, the network is considered converged, and the state variables of the final iteration are output.
[0167] In existing technologies, the inversion process often assumes consistent observation quality. When encountering fluctuations in observations over different periods or unstable local morphological changes, the parameter output is easily influenced by the fluctuation segments and lacks credibility constraints. This invention generates observation confidence based on the temporal changes of respiratory drive characteristics and maps it to the sample weights of the training loss to form a weighted joint loss. At the same time, it uses the correlation criterion between loss reduction and smoothness of state variables to determine the convergence output. The advantage is that it explicitly incorporates observation reliability into the inversion constraints, improves the stability and credibility of outputs such as airway resistance and pleural pressure changes, and reduces the propagation of disturbances from observation fluctuations to the results.
[0168] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
[0169] It should be noted that, in this document, 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 a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0170] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.
[0171] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims
1. A non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception, characterized in that, include: The 4D holographic perception and respiratory observation module receives the echo sequence of the chest and abdomen region from the millimeter-wave sparse aperture array, generates a 4D point cloud sequence of the time-varying morphology of the chest and abdomen surface frame by frame, and constructs a respiratory observation sequence of the chest and abdomen surface. The respiratory observation sequence includes a geodesic length sequence, an area dilatation rate sequence, and a velocity field sequence for the thoracic and abdominal regions. The method for constructing respiratory observation sequences on the thoracic and abdominal surfaces includes: By analyzing the adjacency relationships between different points in each sampling frame, a triangular mesh set of the thoracic and abdominal surface is constructed. Calculate the local principal curvature at each point and converge them to generate a curvature tensor evolution sequence; Construct a set of monitoring anchor point trajectories for the surface regions of the thorax and abdomen, and calculate the geodesic length sequence, area dilation rate sequence, and velocity field sequence of the thorax and abdomen regions. By combining geodesic length sequences, area dilatation rate sequences, and velocity field sequences in the thoracic and abdominal regions, a respiratory observation sequence is generated. The geodesic and curvature energy correlation module generates respiratory drive features by analyzing the correlation between the geodesic length variation attributes and the curvature tensor evolution sequence energy variation attributes. The respiratory drive confidence assessment module obtains the observation confidence based on the temporal changes of the chest and abdominal regions in the respiratory drive characteristics; The viscoelastic dynamics inverse reconstruction module inputs the respiratory drive characteristics into an inverse reconstruction network with nonlinear viscoelastic respiratory dynamics constraints, and outputs a set of mechanical parameters of airway resistance and pleural cavity pressure changes.
2. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 1, characterized in that, The method for generating curvature tensor evolution sequences includes: In each frame, the point set is traversed, and the points obtained during the traversal are marked as target points; Calculate the normal vector of the target point, construct a normal plane coordinate system perpendicular to the normal vector, and construct the edge vector between the target point and each point in the list of neighboring points, denoted as the neighboring edge vector; Calculate the difference between the normal vectors of the neighboring points and the target point, and project the neighboring edge vectors and the difference between the normal vectors onto the normal plane coordinate system to obtain the projection vectors of the neighboring edges and the projection vectors of the difference between the normal vectors. Calculate the outer product matrix of the projection vectors of different neighboring edges separately and sum them to obtain the projection matrix of the neighboring edges; Calculate the product matrix of the projection vector of the neighboring edge formed by the same neighboring point and the normal difference projection vector, and sum the product matrices of different neighboring points to obtain the normal response matrix; The curvature tensor matrix of the target point is calculated based on the projection matrix of the adjacent edge and the normal response matrix. The curvature tensor matrix is then symmetricized and decomposed into eigenvalues to obtain two eigenvalues as the local principal curvatures of the target point. The local principal curvature of each point in the combined point set under different adjacent frames is used to obtain the curvature tensor. The curvature tensors are then spliced together in the order of the frame numbers to obtain the curvature tensor evolution sequence.
3. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 2, characterized in that, The method for generating respiratory observation sequences includes: In each frame's point set, select the triangular meshes located in the thoracic region and the abdominal region respectively, filter the triangular meshes located in the three target regions respectively, and mark them as target meshes; Mark the vertices of different target grids within the target area as anchor points of the target area; For each frame, enumerate the anchor point pairs located in the thoracic region and the abdominal region respectively. For each anchor point pair in the region, calculate all the side lengths of the triangular mesh on the thoracic and abdominal surfaces of each frame. Randomly select one anchor point between the anchor point pairs and start searching for another anchor point along the edge to obtain the search path. Enumerate the search paths to obtain the anchor point pairs; accumulate the lengths of all sides in the anchor point pairs to obtain the path length. For the same pair of anchor points, the path of the anchor point pair with the minimum path length is selected as the geodesic. During the generation of the geodesic, the order of the points passed through is combined to obtain the geodesic path. For each pair of anchor points, the geodesic path is solved repeatedly on consecutive frames, and the path length is accumulated to obtain the geodesic length sequence.
4. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 3, characterized in that, The method for calculating the area expansion rate sequence includes: Using different geodesics on the surface of the chest and abdomen of each frame as the center, geodesic strips are obtained by expanding to both sides with a fixed width. The area of the geodesic strip region is statistically analyzed. For the same anchor point pair in different frames, the difference of the geodesic strip region in adjacent frames is calculated. The area expansion rate is calculated by comparing the difference result with the area of the geodesic strip region in the previous frame in the adjacent frame. The area expansion rates are then combined according to the development order of the frame number to obtain the area expansion rate sequence.
5. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 4, characterized in that, The method for calculating the velocity field sequence in the chest and abdomen region includes: Calculate the displacement vector of each point in adjacent frames, and then calculate the dot product of the unit normal vector and the displacement vector, which is denoted as the normal displacement. The triangular meshes constructed by each point within the set of integration points are recorded as the list of participating meshes for the corresponding points. Calculate the cumulative face value of all triangular meshes in the mesh list, divide the cumulative face value into three equal parts and normalize it to obtain the area weight of the corresponding point; The normal displacements of different points in the surface regions of the thorax and abdomen are fused by area weighting, and the frame interval between the weighted fusion result and the adjacent frames is calculated to obtain the displacement velocity of the thorax and the displacement velocity of the abdomen. By combining the thoracic displacement velocity and the abdominal displacement velocity in the order of development according to the frame number, the velocity field of the thoracic and abdominal regions is obtained.
6. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 5, characterized in that, The method for generating respiratory drive features includes: Identify the monotonic intervals of the geodesic length sequence, and include the monotonic intervals of increasing geodesic length and decreasing monotonic intervals into the candidate sets for the inspiratory and expiratory segments, respectively. The rising and falling slopes of the curvature tensor evolution sequence energy are calculated in the candidate sets of the inspiratory and expiratory segments, respectively, to generate the expansion rate indicator sequence. By analyzing the consistency between the slope change properties in the expansion rate indicator sequence and the growth in the area expansion rate sequence, a set of key moments including the initiation of inhalation, the pause, and the initiation of exhalation is constructed. Extract the duration of inhalation and exhalation from the critical moment set; Extract the maximum displacement velocity of the thoracic and abdominal regions in each respiratory cycle from the velocity field of the thoracic and abdominal regions to obtain the peak expansion rate of the thoracic and abdominal regions. Within each respiratory cycle, the area expansion rate sequence was statistically analyzed, and the area expansion of the thoracic and abdominal regions was obtained and accumulated. The accumulated result was recorded as the body surface expansion proxy amount. By integrating inspiratory duration, expiratory duration, peak expansion rate, inspiratory-expiratory transition steepness, and surface expansion proxy, respiratory drive characteristics of the thoracic and abdominal surface regions were obtained.
7. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 6, characterized in that, The method for generating the expansion rate indication sequence includes: For the thoracic and abdominal regions, the curvature energy is obtained by squaring and summing the local principal curvature of each point in the curvature tensor evolution sequence. The curvature energy of all points within the same region is combined according to the frame number; The curvature energy sequence is divided into interval slices according to the candidate sets of inspiratory and expiratory segments to obtain the energy subsequence sets of inspiratory and expiratory segments respectively. For each inspiratory segment energy subsequence and expiratory segment energy subsequence, sliding fitting is performed within the segment to obtain the rising slope sequence and the falling slope sequence, respectively. The ascending slope sequence and the descending slope sequence are spliced together, and the splicing result is recorded as the expansion rate indicator sequence.
8. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 7, characterized in that, The method for constructing a set of key moments including the initiation of inhalation, the pause, and the initiation of exhalation includes: Identify consecutive non-negative segments and consecutive non-positive segments in the area expansion rate sequence, and match the timestamps corresponding to the consecutive non-negative segments with the timestamps of the rising slope. Mark the earliest time obtained from the matching as the inhalation start time, and the latest time obtained from the matching as the inhalation peak time; Match the timestamps corresponding to consecutive non-positive segments with the timestamps of the descending slope, and mark the earliest matching time as the exhalation start time; By integrating the key moments of inhalation onset, peak inhalation, and exhalation onset, a set of key moments is obtained.
9. The non-contact respiratory mechanics reverse reconstruction system based on 4D holographic field perception according to claim 8, characterized in that, The method for settling the mechanical parameters of output airway resistance and pleural cavity pressure changes includes: The changes in intrapleural pressure and airway resistance are defined as state variables of the network output, and respiratory drive characteristics are used as boundary excitation observations. The breathing drive features and observation confidence are used as boundary excitation inputs to inversely reconstruct the network, and the observation confidence is set as the sample weight of the training loss to generate a weighted joint loss. By analyzing the correlation between the descent rate of the weighted joint loss during the iteration process and the smoothness of the state variable time series, the network convergence is determined and the state variables are output, thus obtaining the set of mechanical parameters.