CN-121641346-B - Infant airway management system and method based on navigation and high-precision dynamic registration

Abstract

The invention discloses an infant airway management system and method based on navigation and high-precision dynamic registration, wherein a personalized airway deformation model is built by utilizing a few sample migration learning network based on physical constraint, an optimal instrument matching scheme is generated, a semantic weighted close-coupled visual inertial odometer is provided, the confidence coefficient of characteristic points is dynamically adjusted by real-time segmentation of airway anatomy semantics, and data are fused under a factor graph optimization framework, so that sub-millimeter level robust positioning is realized. Based on the high-precision track, the system introduces a space-time attention fusion network, captures the long-range time sequence dependency relationship between the operation micro-actions and the physiological parameters, calculates the real-time dynamic risk score, and compensates the airway soft tissue deformation in real time through a non-rigid deformation field model. The invention effectively solves the technical problems of easy loss and risk early warning lag of the existing navigation system in a micro dynamic environment, and obviously improves the safety and success rate of the infant airway management operation.

Inventors

• NI XIN
• ZHAO XIN

Assignees

• 首都医科大学附属北京儿童医院

Dates

Publication Date

20260508

Application Date

20260205

Claims (6)

1. 1. An infant airway management system based on navigation and high-precision dynamic registration, comprising: The modeling and instrument recommendation module is used for generating a personalized 3D airway reference model and instrument recommendation parameters meeting physical prior conditions by utilizing a transfer learning network introducing airway hydrodynamic constraints, wherein the transfer learning network maps sparse image data of an infant to a high-dimensional airway statistical manifold, adopts a conditional variation self-encoder (C-VAE) architecture, receives an infant feature vector comprising age, weight and Cormack-Lehane classification and the sparse image data at the input end, and reconstructs the personalized image data at the output end The device comprises a laryngoscope blade, an airway reference model, a device recommended parameter, a device matching index, a specific calculation mode and a device matching index, wherein the device matching index is used for maximizing the ratio of the outer diameter of the catheter to the narrowest part of the airway, combining with the geometrical factors of the curvature of the laryngoscope blade, ensuring that a preset safety gap is kept between the catheter and the airway, and the device matching index is calculated in a specific calculation mode: Wherein, the Representing the instrument matching index, Indicating the recommended outer diameter of the endotracheal tube, The representation is composed of Airway reference model The effective diameter of the narrowest part of the airway of the infant is measured, Representing laryngoscope blade The matching degree of the curvature of the airway of the infant is used for guiding the laryngoscope to select, Representing weighting coefficients for balancing the importance of catheter size matching and laryngoscope geometry matching, Optimizing targets, finding causes Maximum catheter And laryngoscope blade In combination with each other, Indicating the model of the endotracheal tube to be selected, Representing the laryngoscope blade model to be selected; The position and pose resolving module is used for synchronously acquiring endoscopic images of the laryngoscope and IMU data of the inertial measurement unit, carrying out customized enhancement processing on the endoscopic images of the laryngoscope, carrying out red channel weighted graying processing on the acquired endoscopic images of the laryngoscope so as to highlight mucosa microvascular textures, then applying self-adaptive histogram equalization on the gray images so as to enhance local contrast, extracting microscopic characteristic points of airway mucosa, wherein the microscopic characteristic points comprise mucosa microvascular textures, resolving real-time position and pose and motion trail of the tip of the laryngoscope, resolving the real-time position and pose and motion trail of the tip of the laryngoscope comprises carrying out tight coupling fusion through an error state expansion Kalman filter frame, resolving the real-time six-degree-of-freedom position and motion trail of the tip of the laryngoscope, and solving the problem of positioning drift under the action of quick intubation by fusing high-frequency inertial measurement unit IMU data and frequency vision data through the error state expansion Kalman filter frame, wherein the error state expansion Kalman filter frame maintains an error state vector for estimating and compensating the positioning drift The linear evolution equation is as follows: Wherein the method comprises the steps of For an error state transition matrix, which explicitly describes the coupling effect of zero offset, attitude error of the inertial measurement unit IMU data on position and velocity, The error state vector is represented as a vector of states, In the process, the The noise jacobian matrix is represented by, Representing a system noise vector; The dynamic deformation field registration and compensation module is used for introducing the constraint of a non-rigid deformation field based on the topological correspondence between the micro-feature points and the 3D airway reference model, and calculating and correcting deformation errors generated by respiration or appliance compression in the 3D airway reference model, wherein the micro-feature points and static state in the sparse point cloud of the micro-feature points are matched through local geometric descriptors The non-rigid deformation field describes the displacement from a static 3D model point set to an intraoperative real-time point cloud, and the non-rigid deformation field is resolved through a Coherent Point Drift (CPD) registration algorithm, so that the dynamic compensation of the airway soft tissue is realized on a 3D modeling level; the risk assessment module is used for constructing a multi-mode space-time attention network, taking the characteristics of the motion trail and the real-time physiological signals as input, capturing causal delay characteristics between the operation behaviors and the physiological reactions through a cross attention mechanism, and outputting a real-time dynamic risk score; and the early warning module is used for mapping the corrected 3D airway reference model into a field of view of the mixed reality terminal, enabling the corrected 3D airway reference model to be overlapped with the laryngoscope endoscopic image, and triggering grading early warning according to the real-time dynamic risk score.
2. 2. The system of claim 1, wherein the pose resolving module resolves real-time poses and motion trajectories of the laryngoscope tip further comprises performing close-coupled fusion through a factor graph optimization framework to resolve real-time six-degree-of-freedom poses and motion trajectories of the laryngoscope tip.
3. 3. The system according to claim 2, wherein in the pose resolving module, a factor map optimization framework constructs a joint cost function comprising a priori factor, a pre-integral factor and a semantically weighted visual re-projection factor, and the joint cost function is dynamically modulated and visual re-projection errors are processed to achieve laryngoscope tip positioning.
4. 4. The system of claim 1, wherein to achieve high accuracy positioning, the pose resolving module is configured to incorporate an extrinsic transformation matrix to compensate for physical offset between the laryngoscope camera optical center and inertial measurement unit IMU center at visual observation update, and in combination with zero speed correction (ZUPT) logic, to force zero speed update to suppress integral drift of inertial measurement unit IMU when the laryngoscope is in a static condition.
5. 5. The system of claim 1, wherein a multi-modal spatiotemporal attention network (ST-AFN) in the risk assessment module captures causal delay characteristics by: The track feature encoder is used for extracting high-order kinematic features of the positioning track output by the pose resolving module, and the high-order kinematic features at least comprise jitter and hysteresis entropy; The cross attention module is used for defining the physiological state as Query, the operation track vector comprises a Key vector Key and a Value vector Value, and the attention weight of the operation action feature on the physiological state change is calculated to identify Key operation fragments causing risk rise.
6. 6. An infant airway management method based on navigation and high-precision dynamic registration is characterized by comprising the following steps: Step 1, generating a personalized 3D airway reference model and instrument recommended parameters meeting physical prior conditions by using a transfer learning network introducing airway hydrodynamic constraints, wherein the transfer learning network maps sparse image data of an infant to a high-dimensional airway statistical manifold, adopts a conditional variation self-encoder (C-VAE) architecture, receives an infant feature vector comprising age, weight and Cormack-Lehane classification and the sparse image data at an input end, and reconstructs the personalized image data at an output end The device comprises a laryngoscope blade, an airway reference model, a device recommended parameter, a device matching index, a specific calculation mode and a device matching index, wherein the device matching index is used for maximizing the ratio of the outer diameter of the catheter to the narrowest part of the airway, combining with the geometrical factors of the curvature of the laryngoscope blade, ensuring that a preset safety gap is kept between the catheter and the airway, and the device matching index is calculated in a specific calculation mode: Wherein, the Representing the instrument matching index, Indicating the recommended outer diameter of the endotracheal tube, The representation is composed of Airway reference model The effective diameter of the narrowest part of the airway of the infant is measured, Representing laryngoscope blade The matching degree of the curvature of the airway of the infant is used for guiding the laryngoscope to select, Representing weighting coefficients for balancing the importance of catheter size matching and laryngoscope geometry matching, Optimizing targets, finding causes Maximum catheter And laryngoscope blade In combination with each other, Indicating the model of the endotracheal tube to be selected, Representing the laryngoscope blade model to be selected; Step 2, synchronously acquiring laryngoscope endoscopic images and IMU data of an inertial measurement unit, carrying out customized enhancement processing on the laryngoscope endoscopic images, carrying out red channel weighted graying processing on the acquired laryngoscope endoscopic images so as to highlight mucosa microvascular textures, then applying adaptive histogram equalization on the gray images so as to enhance local contrast, extracting microscopic feature points of airway mucosa, wherein the microscopic feature points comprise mucosa microvascular textures, resolving real-time pose and motion track of a laryngoscope tip, carrying out tight coupling fusion through an error state expansion Kalman filter frame, resolving real-time six-degree-of-freedom pose and motion track of the laryngoscope tip, and solving the problem of positioning drift under the action of quick intubation by fusing high-frequency inertial measurement unit IMU data and low-frequency visual data through the error state expansion Kalman filter frame, wherein in order to estimate and compensate the positioning drift, the error state expansion Kalman filter frame maintains an error state vector The linear evolution equation is as follows: Wherein the method comprises the steps of For an error state transition matrix, which explicitly describes the coupling effect of zero offset, attitude error of the inertial measurement unit IMU data on position and velocity, The error state vector is represented as a vector of states, In the process, the The noise jacobian matrix is represented by, Representing a system noise vector; Step 3, based on the topological correspondence between the micro feature points and the 3D airway reference model, introducing the constraint of a non-rigid deformation field, and calculating and correcting deformation errors generated by respiration or appliance compression in the 3D airway reference model, wherein the micro feature points and static state in the sparse point cloud of the micro feature points are matched through local geometric descriptors The non-rigid deformation field describes the displacement from a static 3D model point set to an intraoperative real-time point cloud, and the non-rigid deformation field is resolved through a Coherent Point Drift (CPD) registration algorithm, so that the dynamic compensation of the airway soft tissue is realized on a 3D modeling level; step 4, constructing a multi-mode space-time attention network, taking the characteristics of the motion trail and the real-time physiological signals as input, capturing causal delay characteristics between the operation behaviors and the physiological reactions through a cross attention mechanism, and outputting real-time dynamic risk scores; And step 5, mapping the corrected 3D airway reference model into a field of view of a mixed reality terminal, enabling the corrected 3D airway reference model to be overlapped with the laryngoscope endoscopic image, and triggering grading early warning according to the real-time dynamic risk score.

Description

Infant airway management system and method based on navigation and high-precision dynamic registration Technical Field The invention belongs to the technical field of medical appliances, and particularly relates to an infant airway management system and method based on navigation and high-precision dynamic registration. Background Infant airway management, particularly tracheal intubation, is a high risk operation in pediatric anesthesia and first aid. The clinical challenges stem from the uniqueness of infant airway anatomy, namely that the effective diameter of the airway is only a few millimeters at critical anatomical sites of small, moist and constant motion (e.g., subglottic stenosis in a three month-old infant). Any minor misoperation may cause serious injury of glottis or subglottal mucosa, and serious complications such as acute edema, laryngeal spasm and even long-term tracheal stenosis are caused. Therefore, the core objective of clinical "prevention of infant glottic and subglottic injuries" requires that the positioning and measurement accuracy of the navigation system must reach the sub-millimeter level, which is a stringent requirement imposed on the prior art. In order to improve the safety and the accuracy of the operation, a Mixed Reality (MR) navigation technology is introduced into the clinic, and aims to accurately superimpose a personalized three-dimensional airway model constructed before the operation in the field of view of an operator in real time. However, when the conventional general navigation technology is applied to a special clinical scene of an infant airway, the technical bottleneck is difficult to surmount, and effective intraoperative real-time spatial registration and dynamic tracking are difficult to realize: 1. The method lacks stable characteristics that the inner wall of the airway is a smooth mucous membrane tissue, and lacks macroscopic geometrical characteristics such as corner points, edges and the like on which the traditional synchronous positioning and mapping (SLAM) algorithm depends, so that the method belongs to a 'no-characteristic' environment macroscopically, and the traditional SLAM algorithm is extremely prone to failure. The invention aims to solve the fundamental problem by creatively taking the mucosa microvascular texture and the soft tissue folds as stable positioning information sources. 2. The dynamic change and non-rigid deformation of the environment, namely spontaneous breathing of the sick children, heartbeat pulsation and contact of an operation instrument can lead to complex non-rigid deformation of airway soft tissues. The existing airway navigation technology mainly relies on rigid registration, and ignores registration errors introduced by severe non-rigid deformation. At the same time, the secretion of the mucous membrane surface and the illumination caused by the reflection of the light source change drastically, further disturbing the stability of the visual features. 3. Real-time and precision bottleneck-the existing general navigation scheme is difficult to meet the severe requirements of operations on real-time when processing complex data of infant airways. The pose update frequency of the navigation system needs to reach the hundred hertz level, and the visual feedback time delay needs to be lower than 100ms. The prior art has significant drawbacks in terms of real-time spatial registration and dynamic tracking. 4. The risk assessment mechanism is split, namely the existing navigation system and the risk assessment system are split, namely the navigation pipeline only and the monitor only. The system lacks a mechanism, and the laryngoscope operation track (such as jitter degree and hysteresis entropy) calculated with high precision is related with physiological parameters of the child patient to carry out causality deduction and real-time risk early warning. In view of the above technical challenges, a brand new technical scheme is urgently needed in the art, and the problems that the existing navigation technology is low in positioning precision, poor in robustness, easy to be subjected to dynamic interference and risk early warning hysteresis can be solved, so that high-precision real-time operation navigation is realized, and safety of clinical operation is guaranteed. The present invention is presented in order to address the above challenges. Disclosure of Invention In order to solve the technical problems, the invention provides an infant airway management system based on navigation and high-precision dynamic registration, which comprises: The modeling and instrument recommendation module is used for generating a personalized 3D airway reference model and instrument recommendation parameters meeting physical priori conditions by utilizing a transfer learning network introducing airway hydrodynamic constraints; the pose resolving module is used for synchronously acquiring the endoscopic image of the laryngoscope and IMU da