Intelligent monitoring and alarming system for displacement of double-lumen bronchial catheter

Abstract

The invention relates to the technical field of displacement monitoring, in particular to an intelligent monitoring and alarming system for displacement of a double-lumen bronchial catheter. The system comprises a catheter positioning information acquisition module, a form reference model construction module, a displacement risk calculation engine module and an intelligent decision-making early warning module, dynamic tracking of three-dimensional space coordinates can be performed by using a three-dimensional space positioning unit, and an original catheter space trajectory data set is generated; constructing an airway-catheter matching form simulation model, and generating a reference catheter-airway relative position relation map; and inputting the original conduit space trajectory data set into the reference conduit-airway relative position relation map to carry out space-time registration and weighted fusion analysis, calculating an instantaneous displacement risk index and an accumulated displacement risk trend at the current moment, generating a dynamic risk decision instruction, outputting an alarm signal and generating differentiated deviation correction prompt information, and sending the alarm log file. According to the invention, the displacement degree, direction and dynamic change rule of the catheter can be clearly presented.

Description

Technical Field

[0001] This invention relates to the field of displacement monitoring technology, and in particular to an intelligent monitoring and alarm system for displacement of a dual-lumen bronchial tube. Background Technology

[0002] Currently, most common methods for monitoring intraoperative displacement of double-lumen endotracheal tubes rely on intermittent fiberoptic bronchoscopy and manual observation. Operators assess tube depth based on experience, auscultate the symmetry of breath sounds in both lungs, and observe changes in airway pressure waveforms to indirectly infer tube position stability. However, existing methods lack individualized airway anatomy references, making it difficult to effectively distinguish between normal physiological displacement and abnormal displacement during dynamic fluctuations in the respiratory cycle. Furthermore, alarm mechanisms rely on single parameter thresholds and exhibit feedback lag, typically triggering alerts only after significant clinical tube displacement has occurred, failing to achieve early prediction and targeted correction of risk trends. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide an intelligent monitoring and alarm system for the displacement of a dual-lumen bronchial tube, in order to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a dual-lumen bronchial tube displacement intelligent monitoring and alarm system includes the following modules:

[0005] The catheter positioning information acquisition module is used to dynamically track the three-dimensional spatial coordinates of the double-lumen bronchial catheter in the insertion state using a three-dimensional spatial positioning unit, collect the real-time position fluctuation signals of the catheter tip and body during the physiological cycle, and generate the original catheter spatial trajectory dataset. The morphological baseline model construction module is used to construct an individualized airway-catheter matching morphological simulation model for the patient based on the patient's preoperative 3D thoracic cavity image and airway anatomical parameters, and generate a baseline catheter-airway relative positional relationship map. The displacement risk calculation engine module is used to input the original catheter spatial trajectory dataset into the baseline catheter-airway relative position relationship map for spatiotemporal registration, calculate the real-time offset vector of the catheter tip relative to the target bronchial opening, the real-time deflection angle of the catheter body relative to the central axis of the main trachea, and the dynamic displacement parameters of the catheter under the physiological motion coupling effect, and generate a displacement risk quantification dataset. The intelligent decision-making and early warning module receives the displacement risk quantification dataset and performs weighted fusion analysis on the real-time offset vector, real-time deflection angle, and duct dynamic displacement parameters. It calculates the instantaneous displacement risk index and cumulative displacement risk trend at the current moment, and generates a dynamic risk decision instruction containing the risk level, displacement direction, and estimated displacement. Upon receiving the dynamic risk decision instruction, when the instantaneous displacement risk index exceeds the preset safety threshold or the cumulative displacement risk trend shows a dangerous slope, it outputs an alarm signal and generates differentiated correction prompt information, and sends an alarm log file containing abnormal coordinates and suggested adjustment vectors.

[0006] The beneficial effects of this invention are: The intelligent monitoring and alarm system for dual-lumen bronchial catheter displacement proposed in this invention consists of a catheter positioning information acquisition module, a morphological reference model construction module, a displacement risk calculation engine module, and an intelligent decision-making and early warning module. Compared with existing technologies, the advantages of this application lie in using a three-dimensional spatial positioning unit to dynamically track the three-dimensional spatial coordinates of the dual-lumen bronchial catheter, collecting positional fluctuation signals of the catheter tip and body during the physiological cycle, generating an original catheter spatial trajectory dataset, and achieving continuous dynamic tracking of the catheter position through the three-dimensional spatial positioning unit. It comprehensively collects the real-time positional fluctuations of the catheter tip and body during the physiological cycle, clearly recording the spatial trajectory changes of the catheter under the influence of physiological activities such as breathing and heartbeat. The generated original catheter spatial trajectory dataset can truly reflect the dynamic displacement characteristics of the catheter. This real-time, continuous positioning and acquisition method effectively avoids the difficulty of capturing subtle dynamic changes by manual observation and the displacement omissions caused by the intermittent nature of fiberoptic bronchoscopy verification, breaking the traditional mode of indirect inference of catheter position. Secondly, based on the patient's preoperative 3D thoracic cavity images and airway anatomical parameters, an individualized airway-catheter matching morphological simulation model is constructed. This model fully incorporates the patient's unique airway anatomy, replicating the airway's shape, diameter, direction, and bronchial opening position to create a personalized simulation model. This model clarifies the relative position of the catheter to the airway in its normal position, and the generated baseline atlas provides a scientific and individualized reference for displacement assessment. This individualized modeling approach can fully adapt to the airway differences among different patients, effectively avoiding displacement assessment biases caused by variations in airway anatomy, clearly defining the boundaries between normal and abnormal displacement within the menstrual cycle, and breaking through the limitations of traditional methods that rely on surgeon experience and have vague reference standards. Then, the original catheter spatial trajectory dataset was spatiotemporally registered with the baseline catheter-airway relative position map. Quantitative calculations were used to generate displacement risk data. Through spatiotemporal registration technology, the real-time spatial trajectory of the catheter was compared with the baseline position to comprehensively calculate the real-time offset vector of the catheter tip relative to the target bronchial opening and the real-time deflection angle of the catheter body relative to the central axis of the main trachea. Simultaneously, the dynamic displacement parameters of the catheter under physiological motion coupling were captured, transforming abstract positional changes into quantifiable and analyzable risk parameters, generating a structured displacement risk dataset. These quantitative parameters clearly present the degree, direction, and dynamic changes of catheter displacement, effectively distinguishing between normal physiological fluctuations and abnormal displacement, avoiding the drawbacks of traditional methods that rely on operator experience and cannot quantify risk. Finally, weighted fusion analysis of the displacement risk dataset was performed to comprehensively calculate the instantaneous displacement risk index and cumulative displacement risk trend, achieving early prediction and dynamic assessment of displacement risk.When risk indicators exceed safety thresholds or show dangerous trends, alarm signals are promptly output, and differentiated corrective prompts are generated. Alarm logs containing abnormal coordinates and suggested adjustment vectors are pushed to provide clear intervention guidance for medical staff. This breaks through the limitations of traditional alarms that only indicate abnormalities without targeted corrective guidance, and greatly improves the timeliness and effectiveness of clinical intervention. Attached Figure Description

[0007] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the modules of the intelligent monitoring and alarm system for dual-lumen bronchial tube displacement of the present invention; Figure 2 for Figure 1 A functional flowchart of the morphological benchmark model construction module; Figure 3 for Figure 1 A functional flowchart of the mid-shift risk calculation engine module. Detailed Implementation

[0008] The technical system of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0009] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0010] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0011] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides an intelligent monitoring and alarm system for dual-lumen bronchial tube displacement, the system comprising the following modules: The catheter positioning information acquisition module is used to dynamically track the three-dimensional spatial coordinates of the double-lumen bronchial catheter in the insertion state using a three-dimensional spatial positioning unit, collect the real-time position fluctuation signals of the catheter tip and body during the physiological cycle, and generate the original catheter spatial trajectory dataset. In this embodiment of the invention, the three-dimensional spatial positioning unit adopts a distributed optical fiber sensor network and is deployed around the outer wall of the patient's trachea through the distributed optical fiber sensor network. The deployment range covers the trachea projection area from the lower edge of the thyroid cartilage to the level of the sternal angle. The optical fiber spacing is controlled at 0.5 mm (exemplary parameter, the optical fiber spacing can be adjusted in the range of 0.3-0.8 mm, and 0.5 mm is preferred in this embodiment). It fits closely to the physiological curvature of the trachea projection area of ​​the neck and upper chest of the human body, ensuring that the monitoring range fully covers the area where the catheter is located. The network transmits low-coherence interference light with a center wavelength of 1310 nm (exemplary parameter; the center wavelength of the low-coherence interference light can be selected from the near-infrared band range of 1200 nm to 1600 nm; 1310 nm is used exemplary in this embodiment). The optical power is controlled at -30 dBm (exemplary parameter; the optical power can be adjusted from -35 dBm to -25 dBm to adapt to different body surface thicknesses). After the interference light penetrates the body surface tissue, it is reflected by the deep tissue and the wall of the double-lumen bronchial duct to form a backscattered signal. The optical fiber receiver synchronously receives the backscattered signal, with the receiving bandwidth set to 100 MHz (exemplary parameter; the receiving bandwidth can be adjusted from 80 MHz to 120 MHz), and the sampling frequency is 1 kHz (exemplary parameter; the sampling frequency can be adjusted from 0.8 kHz to 1.2 kHz). The received backscattered signal is phase demodulated. During demodulation, the optical path difference of the reference arm is adjusted to compensate for the signal phase drift. Then, a wavefront reconstruction algorithm is used to separate the strong reflection signal components generated by the metal reinforcement ring of the duct and the X-ray-proof tip. The amplitude of the strong reflection signal is more than 30 times higher than the amplitude of the background scattering signal (exemplary parameter; the amplitude ratio can be adjusted within the range of 25-35 times to ensure effective screening of characteristic signals). Based on this, the characteristic reflection signal sequence of the duct is obtained. The sequence is subjected to joint time-domain and frequency-domain analysis to extract the optical path difference and Doppler frequency shift corresponding to each signal component. The optical path difference is calculated as ΔL=(λ×Δφ) / (4π), where λ is the center wavelength of the low-coherence interference light, and Δφ is the phase difference of the demodulated signal. The Doppler frequency shift is calculated as f_d=(2v×cosθ) / λ, where v is the velocity of the reflection point, and θ is the angle between the direction of propagation of the interference light and the direction of motion of the reflection point. The axial and radial coordinates of the reflection point are calculated by optical temporal reflection algorithm and optical frequency domain reflection algorithm respectively. The axial coordinate is z=ΔL / 2, and the radial coordinate is r=(c×Δf) / (2×f_s), where c is the speed of light, Δf is the frequency domain peak offset, and f_s is the sampling frequency. The two are combined to generate the spatial coordinate set of the duct scatter point.Based on the temporal continuity of the coordinate set, a continuous spatial curve describing the catheter morphology is obtained through trajectory fitting. 30,000 coordinate data points are collected at a sampling frequency of 10 kHz within a single respiratory cycle (exemplary parameters; the sampling frequency can be adjusted within the range of 8 kHz to 12 kHz, and the number of data points collected in a single respiratory cycle can be adaptively adjusted according to the respiratory rate). The coordinate changes of the catheter tip and three preset main marker points are recorded. The chest wall impedance respiratory waveform and ECG R wave signal are collected simultaneously as time synchronization references. The coordinate change data and physiological signals are aligned with the timestamps to generate the original catheter spatial trajectory dataset.

[0012] The morphological baseline model construction module is used to construct an individualized airway-catheter matching morphological simulation model for the patient based on the patient's preoperative 3D thoracic cavity image and airway anatomical parameters, and generate a baseline catheter-airway relative positional relationship map. In this embodiment of the invention, preoperative thin-slice chest computed tomography (DICOM) image data of the patient is acquired. The image slice thickness is 0.625 mm (exemplary parameter; the image slice thickness can be adjusted within the range of 0.5-1.0 mm to balance imaging accuracy and data processing efficiency). The scanning range is from the skull base to the diaphragm top. A three-dimensional airway cavity model from the glottis to the bilateral main bronchi and lobar bronchi is extracted using a threshold segmentation algorithm. The threshold is set to -500 HU to -100 HU (exemplary parameter; the threshold range can be adjusted within the range of -550 HU to -50 HU to adapt to the differences in airway tissue density among different patients). After segmentation, the internal cavities of the airway cavity are filled using a region growing algorithm. Then, the airway cavity contour is smoothed using a surface reconstruction algorithm. The smoothing iteration count is 10 times (exemplary parameter; the smoothing iteration count can be adjusted within the range of 8-12 times). An individualized three-dimensional geometric model of the airway is generated with a coordinate accuracy of 0.1 mm (exemplary parameter; the coordinate accuracy can be adjusted within the range of 0.08-0.12 mm). Based on the specifications of the double-lumen endotracheal catheter, a three-dimensional solid model of the catheter was retrieved. This model has an outer diameter of 10 mm, an inner diameter of 8 mm, a tip curvature of 15°, a bronchial cuff diameter of 20 mm, a tracheal cuff diameter of 25 mm, and a distance of 30 mm between the two cuffs (these are exemplary parameters and can be flexibly adjusted according to commonly used clinical catheter models; for example, the outer diameter can be in the range of 8-12 mm, and the tip curvature can be in the range of 12°-18°). This model was then matched with an individualized three-dimensional airway geometric model. Based on the expected insertion depth of 25 cm from the anesthesia record (exemplary parameter; the insertion depth can be adjusted within the range of 22-28 cm according to the patient's height and body type) and the information regarding the right side of insertion, the three-dimensional solid model of the catheter was gradually inserted into the right main bronchus along the central path of the airway. Insertion was stopped when the bronchial cuff was 5 mm from the distal end of the opening of the right main bronchus (exemplary parameter; the distance can be adjusted within the range of 3-7 mm to ensure the cuff is in a safe position). The catheter posture was then fixed to generate virtual initial catheter posture data. The Euclidean distance between each point on the catheter surface and the anatomical landmarks on the airway wall is calculated as d = Simultaneously, the spatial angle between the long axis of the duct and the main axis of the airway is calculated. The formula for calculating the angle is cosθ=(a·b) / (|a|×|b|), and the spatial relationship between the duct and the airway is obtained by integrating the results. Simulated physiological movements are applied to the airway model. Respiratory movements are at a frequency of 12 breaths / minute and an expansion / contraction amplitude of 10% of the airway diameter (exemplary parameters; respiratory rate can be adjusted within the range of 10-14 breaths / minute, and expansion / contraction amplitude can be adjusted within the range of 8%-12%). Heartbeat movements are at a frequency of 70 beats / minute and a slight displacement amplitude of 0.5 mm (exemplary parameters; heart rate can be adjusted within the range of 60-80 beats / minute, and displacement amplitude can be adjusted within the range of 0.3-0.7 mm). The allowable elastic deformation and slight movement range of the catheter are calculated, with the deformation coefficient controlled within 0.01 (exemplary parameters; deformation coefficient can be adjusted within the range of 0.008-0.012), the translation range ±1 mm (exemplary parameters; translation range can be adjusted within the range of ±0.8-±1.2 mm), and the rotation range ±5° (exemplary parameters; rotation range can be adjusted within the range of ±4°-±6°). Static data and motion boundary data are fused to generate a baseline catheter-airway relative position relationship map.

[0013] The displacement risk calculation engine module is used to input the original catheter spatial trajectory dataset into the baseline catheter-airway relative position relationship map for spatiotemporal registration, calculate the real-time offset vector of the catheter tip relative to the target bronchial opening, the real-time deflection angle of the catheter body relative to the central axis of the main trachea, and the dynamic displacement parameters of the catheter under the physiological motion coupling effect, and generate a displacement risk quantification dataset. In this embodiment of the invention, the original catheter spatial trajectory dataset is spatiotemporally registered with a reference catheter-airway relative position map. During the registration process, respiratory and electrocardiogram signals are used as time synchronization references to ensure temporal consistency between the trajectory data and the reference map. The real-time coordinates of the catheter tip in the original trajectory data and the ideal tip target coordinates in the reference map are extracted, and the real-time offset vector Δr = (xi - x0, yi - y0, zi - z0) is calculated, where xi, yi, and zi are the real-time tip coordinates, and x0, y0, and z0 are the ideal target coordinates. The offset vector magnitude is calculated as |Δr| = ... The real-time coordinates of three preset marker points on the catheter body are extracted. The real-time principal direction of the catheter body is obtained by solving the eigenvector corresponding to the largest eigenvalue through the covariance matrix. This is compared with the ideal catheter axis direction in the baseline spectrum to calculate the real-time deflection angle in three degrees of freedom. The angle calculation formula is cosθ=(a·b) / (|a|×|b|). Spectral analysis is performed on the catheter tip trajectory, with an analysis frequency range of 0.1-10Hz (exemplary parameter; the analysis frequency range can be adjusted within 0.08-12Hz). The trajectory signal is converted into a frequency domain signal using a Fast Fourier Transform (FFT), with 1024 transformation points (exemplary parameter; the number of transformation points can be adjusted within 512-2048). The power spectral density of the respiratory frequency (fundamental frequency 0.2Hz), the heart rate (fundamental frequency 1.17Hz), and their harmonic frequencies is extracted. The power spectral density calculation formula is PSD=|X(f)|² / T, where X(f) is the frequency domain signal and T is the analysis time. This yields the displacement spectral characteristics of the catheter under the coupling effect of respiration and heartbeat. The real-time offset vector, real-time deflection angle, and displacement spectrum characteristics are synchronously correlated according to the timestamp, and the parameters are analyzed in multiple dimensions to generate a displacement risk quantification dataset (i.e., a multi-dimensional catheter displacement risk quantification dataset).

[0014] The intelligent decision-making and early warning module receives the displacement risk quantification dataset and performs weighted fusion analysis on the real-time offset vector, real-time deflection angle, and duct dynamic displacement parameters. It calculates the instantaneous displacement risk index and cumulative displacement risk trend at the current moment, and generates a dynamic risk decision instruction containing the risk level, displacement direction, and estimated displacement. Upon receiving the dynamic risk decision instruction, when the instantaneous displacement risk index exceeds the preset safety threshold or the cumulative displacement risk trend shows a dangerous slope, it outputs an alarm signal and generates differentiated correction prompt information, and sends an alarm log file containing abnormal coordinates and suggested adjustment vectors.

[0015] In this embodiment of the invention, by receiving a displacement risk quantification dataset, the real-time offset vector sequence, real-time deflection angle sequence, and displacement spectrum feature data are extracted, and a weighted fusion analysis structure is constructed to process the three sets of data. First, the data is standardized using the formula x'=(x-μ) / σ, where x is the original data, μ is the data mean, and σ is the data standard deviation. Then, feature encoding is used to transform it into a feature vector of a unified dimension. A time-sliding analysis module is introduced, setting 10 sampling times as a time window (exemplary parameter; the time window can be adjusted within the range of 8-12 sampling times) to capture temporal correlation patterns. The instantaneous displacement risk index is calculated through weighted fusion operations, with weights allocated as follows: offset vector amplitude 0.3, direction persistence 0.2, deflection angle stability 0.2, and non-physiological frequency band spectrum energy concentration 0.3 (exemplary weight allocation; can be fine-tuned within ±0.05 according to clinical needs). The risk index calculation formula is RI=0.3×|Δr| + 0.2×D+ 0.2×S + 0.3×E, with an index range of 0 to 10. The instantaneous risk index is integrated over 10 seconds using the formula ∫RI(t)dt (t ranges from 0 to 10 seconds). This integration, combined with historical risk data, extrapolates the risk trend and generates a cumulative displacement risk trend line. Based on the instantaneous risk index range and the trend line slope, multi-level risk assessment logic is triggered, generating a dynamic risk decision instruction containing risk level, displacement direction, and estimated displacement. The estimated displacement is calculated as Δs=v×t, where v is the average displacement velocity over the past 5 seconds and t is 10 seconds. When the instantaneous risk index exceeds a preset threshold or the trend line exhibits a dangerous slope, three alarm units are activated simultaneously. Differentiated audio-visual and tactile alarms are generated based on the risk level. A formatted alarm message is generated, containing the patient identifier, alarm timestamp, abnormal coordinates, and a suggested adjustment vector. The suggested adjustment vector is calculated as Δr'=-k×Δr, where k is the adjustment coefficient, which is 0.5 for low, 0.8 for medium, and 1.0 for high risk (example adjustment coefficients, which can be fine-tuned according to clinical intervention needs). An alarm message is pushed out, and a structured alarm log file is stored.

[0016] Furthermore, the three-dimensional spatial positioning unit adopts a distributed optical fiber sensor network, which is deployed around the outer wall of the patient's trachea and arranged in a ring array.

[0017] In this embodiment of the invention, the three-dimensional spatial positioning unit employs a distributed optical fiber sensor network. This network is specifically designed for the dynamic tracking of the catheter's three-dimensional coordinates. It is deployed by surrounding the outer wall of the patient's trachea in a ring array, ensuring comprehensive coverage of the tracheal projection area where the catheter is located. Specific deployment details: The ring array of the distributed optical fiber sensor network conforms to the physiological curvature of the tracheal projection area in the neck and upper chest, covering the tracheal projection area from the lower edge of the thyroid cartilage to the level of the sternal angle. The fiber spacing in the ring array is controlled at 0.5 mm (adjustable within the range of 0.3-0.8 mm), with adjacent fibers arranged parallel to each other, conforming to the skin surface without affecting the patient's normal physiological activities. The optical fibers are made of flexible material, allowing for slight deformation with the patient's neck movements, preventing positional shifts due to changes in body position and ensuring monitoring stability. The transmitter and receiver of the ring array are integrated into the same monitoring terminal, facilitating centralized signal processing and transmission, and enabling real-time acquisition and analysis of the catheter's three-dimensional coordinates.

[0018] Furthermore, the catheter positioning information acquisition module includes the following functions: The distributed optical fiber sensing network is based on a preset physiological curvature and fits onto the patient's corresponding neck and upper chest tracheal projection area. It emits detection light and receives backscattered signals formed by reflections from deep tissues and the duct wall. The backscattered signals are processed by phase demodulation and wavefront reconstruction to separate the strong reflection signal components generated by the corresponding feature markers of the double-lumen bronchial duct, thus obtaining the duct feature reflection signal sequence. The time and frequency domains of the catheter characteristic reflection signal sequence are jointly analyzed to extract the characteristic parameters corresponding to each signal component. The preliminary coordinates of each reflection point in three-dimensional space are calculated by the optical domain reflection algorithm to generate a catheter scatter point spatial coordinate set. Based on the temporal continuity of the catheter scatter point spatial coordinate set, the trajectory of the key marker points on the catheter tip region and the catheter body is fitted to generate a continuous spatial curve describing the catheter morphology. The continuous spatial curve is sampled at high frequency within a single respiratory cycle. The coordinate changes of the catheter tip and multiple preset main marker points in the three-dimensional coordinate system over time are recorded. The physiological phase marker signal of this period is collected synchronously as a time synchronization reference to generate an original catheter spatial trajectory dataset with physiological phase markers. The sampling frequency of the high-frequency sampling is not less than 1 kHz.

[0019] In this embodiment of the invention, the distributed optical fiber sensing network is arranged to fit the physiological curvature of the trachea projection area of ​​the neck and upper chest of the human body. The fiber spacing of the ring array is controlled at 0.5 mm (exemplary parameter, adjustable within the range of 0.3-0.8 mm). The fitting range covers the trachea projection area from the lower edge of the thyroid cartilage to the level of the sternal angle. The ring array emits low-coherence interference light with a center wavelength of 1310 nm (exemplary parameter, adjustable within the range of 1200 nm to 1600 nm near-infrared band), and the optical power is controlled at -30 dBm (exemplary parameter, adjustable within the range of -35 dBm to -25 dBm). After the interference light penetrates the body surface tissue, it is reflected by the deep tissue and the wall of the double-lumen bronchus duct to form a backscattered signal. The optical fiber receiving end of the ring array synchronously receives the backscattered signal, and the receiving bandwidth is set to 100 MHz (exemplary parameter, adjustable within the range of 80 MHz to 120 MHz), and the sampling frequency is 1 kHz (exemplary parameter, adjustable within the range of 0.8 kHz to 1.2 kHz). The received backscattered signal is demodulated using a phase-generated carrier demodulation method. During demodulation, the optical path difference of the reference arm is adjusted to compensate for the signal phase drift. Then, a Fourier transform wavefront reconstruction algorithm is used to reconstruct the wavefront of the demodulated signal, separating the strong reflection signal components generated by the metal reinforcing ring and the radiopaque tip of the double-lumen bronchial tube. The amplitude of the strong reflection signal is more than 30 times higher than the amplitude of the background scattering signal (exemplary parameter, adjustable within the range of 25-35 times). Based on this, the characteristic reflection signal sequence of the tube is obtained. The duration of the signal sequence is 10 seconds (exemplary parameter, adjustable within the range of 8-12 seconds), containing 2000 signal sampling points (exemplary parameter, adaptively adjustable according to the sampling frequency). A joint time-domain and frequency-domain analysis was performed on the characteristic reflection signal sequence of the duct. The time-domain analysis adopted the sliding window method with a window length of 50 sampling points (exemplary parameter, adjustable within the range of 40-60 sampling points) and an overlap rate of 50% (exemplary parameter, adjustable within the range of 40%-60%). The optical path difference corresponding to each signal component was extracted. The frequency-domain analysis adopted the fast Fourier transform with 1024 transform points (exemplary parameter, adjustable within the range of 512-2048). The Doppler frequency shift corresponding to each signal component was extracted. The optical path difference was calculated using the formula ΔL=(λ×Δφ) / (4π), where λ is the center wavelength of the low-coherence interference light, Δφ is the phase difference of the demodulated signal, and the Doppler frequency shift was calculated using the formula f_d=(2v×cosθ) / λ, where v is the velocity of the reflection point and θ is the angle between the direction of propagation of the interference light and the direction of motion of the reflection point.The axial coordinates of each reflection point are calculated using an optical temporal reflectance algorithm, which utilizes the linear relationship between optical path difference and axial distance. The axial coordinate is z = ΔL / 2. The radial coordinates of each reflection point are calculated using an optical frequency domain reflectance algorithm, which utilizes the correspondence between the peak position of the frequency domain signal and the radial distance. The radial coordinate is r = (c × Δf) / (2 × f_s), where c is the speed of light, Δf is the frequency domain peak offset, and f_s is the sampling frequency. Combining the axial and radial coordinates, the preliminary coordinates of each reflection point in three-dimensional space are generated, which in turn generates a scatter point spatial coordinate set for the conduit. The coordinate set contains 1000 discrete points (exemplary parameters, adjustable within the range of 800-1200 discrete points). Based on the temporal continuity of the scattered spatial coordinate set of the catheter, cubic spline interpolation is used to fit the trajectory of the scattered points of key markers on the catheter tip and the catheter body. During the fitting process, time is used as the independent variable and coordinates as the dependent variable. The interpolation function coefficients are solved using the least squares method to ensure that the deviation between the fitted curve and the scattered points is controlled within 0.1 mm (exemplary parameter, adjustable within the range of 0.08-0.12 mm), generating a continuous spatial curve describing the catheter morphology. High-frequency sampling is performed on the continuous spatial curve within a single respiratory cycle, with the sampling frequency increased to 10 kHz (exemplary parameter, adjustable within the range of 8 kHz to 12 kHz). A single respiratory cycle is calculated as 3 seconds, and a total of 30,000 coordinate data points are collected (exemplary parameter, adaptively adjustable according to respiratory rate). The coordinate changes of the catheter tip and three preset main body markers in the three-dimensional coordinate system over time are recorded. The three preset main body markers are located at 10 cm, 20 cm, and 30 cm proximal to the catheter, respectively (exemplary parameter, marker position can be adjusted according to catheter length). Thoracic impedance respiratory waveforms and ECG R-wave signals are simultaneously acquired during this period as time synchronization references. Thoracic impedance respiratory waveforms were acquired using electrode pads attached to both sides of the thoracic cavity, with an impedance measurement range of 100-1000Ω (exemplary parameter, adjustable within the range of 80-1200Ω). ECG R-wave signals were acquired using standard limb leads, with a signal amplification factor of 1000x (exemplary parameter, adjustable within the range of 800-1200x) and a filtering frequency range of 0.5-50Hz (exemplary parameter, adjustable within the range of 0.3-60Hz). The coordinate change data were aligned with the respiratory and ECG signals by timestamps to generate a raw catheter spatial trajectory dataset with physiological phase markers. Each coordinate data point corresponds to a respiratory phase marker and an ECG phase marker.

[0020] Furthermore, as an embodiment of the present invention, reference is made to... Figure 2 As shown, Figure 1 A functional flowchart of the morphological baseline model construction module is shown in this embodiment. The morphological baseline model construction module includes the following functions: Import the patient's preoperative thin-section chest computed tomography (DICOM) image data and extract the three-dimensional airway cavity model from the glottis to the bilateral main bronchi and lobar bronchi to generate an individualized three-dimensional airway geometric model. In the individualized three-dimensional airway geometric model, according to the model and specification parameters of the double-lumen bronchial tube, call the catheter three-dimensional entity database to match the corresponding model of the catheter three-dimensional entity model. Based on the expected insertion depth and lateral information of the catheter in the anesthesia record, a virtual insertion operation is performed in the individualized three-dimensional geometric model of the airway. The three-dimensional solid model of the catheter is inserted into the target main bronchus according to the clinical operation specifications, and the initial pose data of the virtual catheter is generated. Based on the initial pose data of the virtual catheter, the spatial relationship between the catheter and the airway is simulated under ideal alignment, and the normal physiological motion envelope boundary of the catheter is defined. The spatial relationship between the catheter and the airway and the normal physiological motion envelope boundary of the catheter are integrated to generate a baseline relative positional relationship map of the catheter and the airway.

[0021] In this embodiment of the invention, preoperative thin-slice chest computed tomography (DICOM) image data of the patient is acquired. The image slice thickness is 0.625 mm (exemplary parameter, adjustable within the range of 0.5-1.0 mm). The scanning range extends from the base of the skull to the top of the diaphragm. A threshold segmentation algorithm is used to extract a three-dimensional airway cavity model from the glottis to the bilateral main bronchi and lobar bronchi. The threshold is set to -500 HU to -100 HU (exemplary parameter, adjustable within the range of -550 HU to -50 HU). After segmentation, the internal cavities of the airway cavity are filled using a region growing algorithm. Then, a surface reconstruction algorithm is used to smooth the airway cavity contour. The smoothing iteration count is 10 times (exemplary parameter, adjustable within the range of 8-12 times). An individualized three-dimensional geometric model of the airway is generated. The model contains the three-dimensional coordinate information of the airway wall, with a coordinate accuracy of 0.1 mm (exemplary parameter, adjustable within the range of 0.08-0.12 mm). In the personalized airway three-dimensional geometric model, the three-dimensional solid model of the catheter is retrieved according to the model and specification parameters of the double-lumen bronchial catheter. The database stores three-dimensional solid models of different models of catheters. Based on the parameters of the catheter's outer diameter of 10 mm, inner diameter of 8 mm, tip curvature of 15°, bronchial cuff diameter of 20 mm, tracheal cuff diameter of 25 mm, and distance between the two cuffs of 30 mm (the above are all exemplary parameters and can be adjusted according to commonly used clinical models), the three-dimensional solid model of the corresponding model of the catheter is matched. This model includes the three-dimensional geometric dimensions and relative positional relationships of each part of the catheter. Based on the expected insertion depth of 25 cm (exemplary parameter, adjustable within the range of 22-28 cm) and the information that the side is right, a virtual insertion operation is performed in a personalized airway 3D geometric model. Using coordinate translation and rotation algorithms, the tip of the 3D solid model of the catheter is gradually inserted into the right main bronchus from the glottic inlet along the central path of the airway. During the insertion process, the distance between the bronchial cuff and the opening of the right main bronchus is calculated in real time. Insertion is stopped when the cuff is 5 mm away from the distal end of the opening of the right main bronchus (exemplary parameter, adjustable within the range of 3-7 mm). The pose of the 3D solid model of the catheter is fixed, and the initial pose data of the virtual catheter is generated. The pose data includes the 3D coordinates and attitude angle information of each discrete point of the catheter. Based on the initial pose data of the virtual catheter, the spatial relationship of the catheter in an ideal alignment state is simulated. A spatial distance calculation algorithm is used to calculate the shortest spatial distance between each point on the catheter tip, cuff, and catheter body surface and each anatomical landmark on the airway wall. Anatomical landmarks include the glottis, carina, right main bronchus opening, and left main bronchus opening. The shortest spatial distance is obtained by calculating the Euclidean distance between the two points, d = ... Where (x1, y1, z1) are the coordinates of points on the catheter surface, and (x2, y2, z2) are the coordinates of anatomical landmarks on the airway wall. At the same time, the spatial angle between the long axis of the catheter and the main axis of the airway is calculated. The spatial angle is obtained by calculating the angle between the direction vectors of the two axes. The calculation formula is cosθ=(a·b) / (|a|×|b|), where a is the direction vector of the long axis of the catheter, b is the direction vector of the main axis of the airway, and θ is the spatial angle between the two axes. By integrating the above distance and angle data, the spatial relationship between the catheter and the airway is generated. Simulated physiological breathing and heartbeat movements are applied to a personalized 3D airway geometric model. The respiratory motion simulation uses a sine curve to control the expansion and contraction of the airway model, with an expansion amplitude of 10% of the airway diameter (exemplary parameter, adjustable within the range of 8%-12%) and a respiratory rate of 12 breaths / minute (exemplary parameter, adjustable within the range of 10-14 breaths / minute). The heartbeat motion simulation uses a cosine curve to control slight displacement of the airway model, with a displacement amplitude of 0.5 mm (exemplary parameter, adjustable within the range of 0.3-0.7 mm) and a heart rate of 70 beats / minute (exemplary parameter, adjustable within the range of 60-80 beats / minute). The allowable elastic deformation and micro-translational / rotation range of the virtual catheter's initial pose data under simulated motion are calculated. The elastic deformation is controlled by a finite... The meta-analysis algorithm calculates the deformation coefficient, controlling it within 0.01 (exemplary parameter, adjustable within the range of 0.008-0.012), the micro-translation range is ±1 mm (exemplary parameter, adjustable within the range of ±0.8-±1.2 mm), and the rotation range is ±5° (exemplary parameter, adjustable within the range of ±4°-±6°). This defines the normal physiological motion envelope boundary of the catheter. The spatial relationship between the catheter and the airway is fused with the physiological motion envelope boundary data. During the fusion process, a weighted average method is used, with a static data weight of 0.7 and a motion boundary data weight of 0.3 (exemplary weight, fine-tunable within the range of ±0.05). A baseline catheter-airway relative positional relationship map is generated, which includes the omnidirectional relative pose parameters of the catheter and airway in both static and dynamic states.

[0022] Furthermore, the simulated catheter-airway spatial relationship under ideal alignment includes: From the initial pose data of the virtual catheter and the individualized three-dimensional geometric model of the airway, discrete point cloud datasets representing the geometric features of the model surface are extracted respectively. Based on the discrete point cloud datasets, the Euclidean distance from each discrete point on the catheter surface to all discrete points on the inner wall of the airway is calculated. The minimum distance value corresponding to each point on the catheter surface and its corresponding nearest point coordinates on the inner wall of the airway are selected to generate a preliminary point-to-point distance mapping table. The initial point-to-point distance mapping table is filtered to remove non-contact abnormal long-distance point pairs, and point pairs that are suspected to represent potential contact or proximity relationships are retained to generate a dataset of effective contact and proximity point pairs. Based on the effective contact proximity point pair dataset, the direction vectors of the catheter's long axis and the airway's main axis are obtained, and the three-dimensional spatial angle between them is calculated. The projection components of this angle on the coronal and sagittal planes are also calculated to generate the spatial angle data of the catheter-airway axis. The distance statistics parameters of all point pairs in the effective contact proximity point pair dataset are integrated with the spatial angle data of the catheter-airway axis to generate the spatial relationship of the catheter-airway under ideal alignment.

[0023] In this embodiment of the invention, the three-dimensional coordinates of 100,000 discrete points on the surface of the 3D solid model of the catheter are extracted from the initial pose data of the virtual catheter (exemplary parameters, adjustable within the range of 80,000-120,000 discrete points). These discrete points are uniformly distributed on the catheter surface, with an adjacent point spacing of 0.1 mm (exemplary parameters, adjustable within the range of 0.08-0.12 mm). Simultaneously, the three-dimensional coordinates of 100,000 discrete points on the surface of the airway inner wall are extracted from the individualized airway 3D geometric model (exemplary parameters, adjustable within the range of 80,000-120,000 discrete points). These discrete points are also uniformly distributed on the airway inner wall, with an adjacent point spacing of 0.1 mm (exemplary parameters, adjustable within the range of 0.08-0.12 mm). These two sets of coordinate data are then organized into a discrete point cloud dataset on the model surface. Based on this discrete point cloud dataset, a double-loop traversal algorithm is used to traverse all discrete points on the airway inner wall for each discrete point on the catheter surface, calculating the Euclidean distance d between them. Where (x1, y1, z1) are the coordinates of discrete points on the catheter surface, and (x2, y2, z2) are the coordinates of discrete points on the airway inner wall. The minimum distance value corresponding to each point on the catheter surface and its corresponding nearest point coordinate on the airway inner wall are selected. The minimum distance value and nearest point coordinate of all points on the catheter surface are associated with their own coordinates to generate a preliminary point-to-point distance mapping table. The mapping table contains 100,000 point-to-point distance records (exemplary parameters, which can be adjusted according to the number of discrete points). The initial point-to-point distance mapping table is filtered using a Gaussian filtering algorithm. The filter window size is 5×5 (exemplary parameter, adjustable within the range of 4×4-6×6), and the standard deviation is 0.5 (exemplary parameter, adjustable within the range of 0.4-0.6). Non-contact abnormal long-distance point pairs caused by airway bifurcation or folds are removed. Abnormal long-distance point pairs have a distance value greater than the average of all distance values ​​plus 3 times the standard deviation. Point pairs with a distance value less than or equal to the average plus 3 times the standard deviation are retained. These point pairs represent potential contact or proximity relationships, generating a dataset of effective contact proximity point pairs. The dataset contains 80,000 effective point pair records (exemplary parameter, adjustable according to the number of records in the initial mapping table). Based on the effective contact proximity pair dataset, a vector fitting algorithm was used to calculate the direction vector of the catheter's long axis. The coordinates of the catheter tip (x0, y0, z0) and the coordinates of the carina (x1, y1, z1) were selected. The direction vector of the catheter's long axis points from the catheter tip to the carina. The vector calculation formula is a = (x1 - x0, y1 - y0, z1 - z0). At the same time, a path fitting algorithm was used to extract the central path from the glottic coordinates (x2, y2, z2) to the target main bronchus terminal coordinates (x3, y3, z3). This central path was used as the airway main axis, and the direction vector of the airway main axis was calculated as b = (x3 - x2, y3 - y2, z3 - z2), generating axis direction vector data. The three-dimensional spatial angle between the duct's long axis direction vector and the airway's main axis direction vector is calculated based on the axial direction vector data. The formula for calculating the angle is cosθ=(a·b) / (|a|×|b|), where a·b is the dot product of the two vectors, |a| is the magnitude of the duct's long axis direction vector, and |b| is the magnitude of the airway's main axis direction vector. The three-dimensional spatial angle θ is obtained by solving the inverse cosine function. Then, the projection components of this angle in the coronal and sagittal planes are calculated. The coronal projection component θx=θ×cosα, and the sagittal projection component θy=θ×sinα, where α is the angle between the duct's long axis and the coronal plane, thus generating the duct-airway axis spatial angle data.The distance values ​​of all point pairs in the effective contact proximity point pair dataset are integrated, and the arithmetic mean method is used to calculate the average distance. The calculation formula is μ=(d1+d2+...+dn) / n, where d1 to dn are the distance values ​​of all effective point pairs, and n is the number of effective point pairs. The standard deviation of the distance is calculated using the standard deviation formula σ=√[Σ(Di-μ)² / n], where Di is the distance value of a single effective point pair. At the same time, the maximum value dmax and the minimum value dmin among the distance values ​​are selected. Combined with the spatial angle data of the catheter-airway axis, the average distance, standard deviation, maximum and minimum values ​​are integrated with the spatial angle of the axis, and the projection components of the coronal and sagittal planes to generate the spatial relationship of the catheter-airway under ideal alignment.

[0024] Furthermore, the filtering process for the initial point-to-point distance mapping table includes: Obtain the distance values ​​recorded in the preliminary point-to-point distance mapping table, draw a histogram of the statistical distribution of all distance values, and set a dynamic filtering threshold based on the distance percentile according to the distribution characteristics; compare the distance values ​​between each pair of points with the dynamic filtering threshold, and filter out a list of abnormal point pairs to be removed; For each point pair in the list of abnormal point pairs to be removed, spatial position relationship analysis is performed, invalid neighboring point pairs are removed, and point pairs near the expected contact area of ​​the catheter are retained and subjected to secondary discrimination. By integrating the valid points after secondary discrimination, a dataset of valid contact neighbor pairs is generated.

[0025] In this embodiment of the invention, the Euclidean distance values ​​between each pair of catheter surface points and airway inner wall points in 100,000 records of the preliminary point-to-point distance mapping table are extracted (exemplary parameters, which can be adjusted according to the number of records in the preliminary mapping table). A histogram drawing algorithm is used to draw a statistical distribution histogram of all distance values. The horizontal axis of the histogram is the distance value, ranging from 0 to 10 mm (exemplary parameter, which can be adjusted within the range of 0-12 mm), with an interval of 0.1 mm (exemplary parameter, which can be adjusted within the range of 0.08-0.12 mm). The vertical axis is the number of point pairs corresponding to the distance value. By analyzing the distribution characteristics of the distance values ​​through the histogram, the distance values ​​show a normal distribution trend. A dynamic filtering threshold based on the distance percentile is set, and the 95th percentile of the distance value is taken as the dynamic filtering threshold (exemplary parameter, which can be adjusted within the range of 90%-98%). The calculation method is to sort all distance values ​​from smallest to largest, and select the 95,000th distance value after sorting as the threshold (exemplary parameter, which can be adjusted according to the total number of records). At the same time, the mean μ, standard deviation σ, and median of all distance values ​​are calculated to generate statistical feature data of distance distribution. Based on the statistical characteristics of distance distribution, the distance value between each pair of points is compared with the dynamic filtering threshold one by one. The comparison adopts the difference comparison method, and the difference between the distance value and the dynamic filtering threshold is calculated as Δd = d - threshold. If Δd > 0, that is, the distance value is greater than the dynamic filtering threshold, the point pair is determined to be a non-contact abnormal point pair caused by the gap of the anatomical structure. The coordinate association identifier of the point pair is recorded, and a list of abnormal point pair identifiers to be removed is generated. The list contains 12,000 abnormal point pair identifiers (exemplary parameters, which can be adjusted according to the number of records in the initial mapping table). For each point pair in the list of abnormal point pairs to be removed, a spatial location algorithm is used to analyze its spatial relationship, extract the three-dimensional coordinates of the duct surface point and the corresponding nearest airway inner wall point, and combine the airway partition information in the individualized airway three-dimensional geometric model to determine the position of the nearest airway inner wall point. If the x-coordinate of the nearest airway inner wall point is less than the x-coordinate of the carina center and the duct insertion side is the right side, that is, it is located in the contralateral bronchus, or if the z-coordinate of the nearest airway inner wall point is greater than the z-coordinate of the end of the target main bronchus, that is, it is located in the lobar bronchus far from the main airway, then the point pair is confirmed as an invalid neighboring point pair, and its identifier is added to the list of invalid spatially associated point pairs. The list contains 8000 invalid point pair identifiers (example parameters, which can be adjusted according to the number of records in the list to be removed).In the list of abnormal point pairs to be removed, the remaining 4000 point pairs are extracted (exemplary parameters, which can be adjusted according to the number of records in the list to be removed and the invalid list). A region matching algorithm is used to determine whether they are near the expected contact area of ​​the catheter. The expected contact area of ​​the catheter is the airway segment from the glottis to the opening of the target main bronchus, and its three-dimensional coordinate range is x∈[100,150] mm, y∈[50,80] mm, z∈[200,300] mm (exemplary parameters, which can be adjusted according to the patient's airway size). If the coordinates of the catheter surface point in the point pair are within this range, it is determined that it is near the expected contact area, and the expected contact area point pair screening result is generated, containing 3500 point pair identifiers (exemplary parameters, which can be adjusted according to the number of remaining point pairs). 3500 point pairs that are not in the list of invalid spatial association point pairs and are located near the expected contact area are reclassified as point pairs to be reviewed (example parameters, which can be adjusted according to the screening results). The curvature calculation algorithm is used to calculate the local curvature of the airway at the location of the point pair to be reviewed. The curvature calculation formula is k=|d²r / ds²|, where r is the position vector of the airway wall curve and s is the arc length of the curve. If the local curvature of the airway is greater than 0.05 mm. -1 (Example parameters, ranging from 0.04 to 0.06 mm) -1 If a point pair is found to be within the range of 0.4-0.6 mm and its distance is less than the dynamic filtering threshold plus 0.5 mm (exemplary parameter, adjustable within the range of 0.4-0.6 mm), it is considered a valid point pair, generating a secondary valid point pair list containing 3200 point pair identifiers (exemplary parameter, adjustable according to the number of point pairs to be reviewed). All point pairs in the invalid spatial association point pair list are removed from the initial point-to-point distance mapping table, and the point pairs in the secondary valid point pair list are added to the filtered point pair set. The resulting dataset of valid contact neighbor point pairs contains 85200 valid point pair records (exemplary parameter, adjustable according to the number after integration).

[0026] Furthermore, as an embodiment of the present invention, reference is made to... Figure 3 As shown, Figure 1 A functional flowchart of the mid-shift risk calculation engine module is shown. In this embodiment, the shift risk calculation engine module includes the following functions: The system receives the original catheter spatial trajectory dataset with physiological phase markers and uses the catheter-airway spatial relationship and the normal physiological motion envelope boundary of the catheter in the baseline catheter-airway relative position relationship map as the registration and calculation benchmark; it performs temporal segmentation on the original catheter spatial trajectory dataset to generate a temporalized trajectory segment dataset. For each temporal trajectory segment, calculate the instantaneous offset vector sequence of the catheter tip's real-time coordinates and the ideal tip target coordinates, as well as the real-time deflection angle sequence of the catheter body; Spectral analysis of the catheter tip's motion trajectory is performed to extract characteristic parameters of the trajectory signal at physiological frequencies and their harmonic frequencies, thereby generating dynamic displacement parameters of the catheter. The instantaneous offset vector sequence of the tip, the real-time deflection angle sequence, and the dynamic displacement parameters of the catheter are synchronously fused according to the timestamp, and multi-parameter correlation analysis is performed to generate a displacement risk quantification dataset.

[0027] In this embodiment of the invention, a raw catheter spatial trajectory dataset with physiological phase markers is received. This dataset contains 30,000 coordinate data points and corresponding physiological phase markers (exemplary parameters, which can be adjusted according to the sampling frequency and acquisition time). At the same time, the spatial relationship between the catheter and the airway and the physiological motion envelope boundary in the baseline catheter-airway relative position relationship map are retrieved as the reference for registration and calculation. A temporal segmentation algorithm is used to segment the original catheter spatial trajectory dataset temporally. By identifying the peak and trough values ​​of the thoracic impedance respiratory waveform, the end-expiratory and end-inspiratory time points are determined. Using these two time points as boundaries, the continuous trajectory data is segmented into trajectory segments within a single respiratory cycle. A single respiratory cycle is calculated as 3 seconds (exemplary parameter, adjustable within the range of 2.5-3.5 seconds), resulting in a total of 10 trajectory segments (exemplary parameter, adjustable according to the total acquisition time). Each segment contains 3000 coordinate data points (exemplary parameter, adjustable according to the segment length and sampling frequency). Based on the ECG R-wave markers, each respiratory cycle is further segmented into cardiac cycle-related subphases, using the time interval between two adjacent R waves as boundaries. Each respiratory cycle contains 5 subphases (exemplary parameter, adjustable according to the heart rate), generating a temporally phased trajectory segment dataset containing 50 temporally phased trajectory segments (exemplary parameter, adjustable according to the number of trajectory segments and subphases). For each temporally phased trajectory segment, a spatial alignment algorithm is used to spatially align the real-time coordinates of the duct tip with the ideal tip target coordinates defined in the baseline duct-airway relative position relationship map. The ideal tip target coordinates are (x0, y0, z0), and the real-time coordinates are (xi, yi, zi). The offset vector Δr = (xi-x0, yi-y0, zi-z0) between the two is calculated. The real-time coordinates of all 300 sampling times within the trajectory segment are traversed (exemplary parameters, which can be adjusted according to the number of data points in the segment) to generate a sequence of instantaneous tip offset vectors. For each temporal trajectory segment, the real-time coordinates of three preset marker points on the catheter body are extracted (exemplary parameters, 2-4 marker points can be set). The real-time principal direction of the spatial curve formed by these three points is calculated using a principal direction extraction algorithm. The principal direction vector is obtained by calculating the covariance matrix of the coordinates of the three points and solving for the eigenvector corresponding to the largest eigenvalue of the covariance matrix. The real-time principal direction vector is compared with the ideal catheter axis direction vector defined in the baseline catheter-airway relative position relationship map, and the spatial angle between the two is calculated. The angle calculation formula is cosθ=(a·b) / (|a|×|b|), thus obtaining the real-time deflection angle sequence of the catheter body in the three degrees of freedom of x, y, and z.A spectrum analysis algorithm was used to perform spectrum analysis on the catheter tip motion trajectory in the original catheter spatial trajectory dataset. The analysis frequency range was 0.1-10Hz (exemplary parameter, adjustable within the range of 0.08-12Hz). The trajectory signal was converted into a frequency domain signal using Fast Fourier Transform (FFT) with 1024 transform points (exemplary parameter, adjustable within the range of 512-2048). The power spectral density of the trajectory signal was extracted at the fundamental respiratory rate of 12 breaths / min (0.2Hz), the fundamental heart rate of 70 beats / min (1.17Hz), and their harmonic frequencies. The power spectral density was calculated using the formula PSD=|X(f)|² / T, where X(f) is the frequency domain signal and T is the analysis time. The amplitude-frequency characteristics of the trajectory oscillation under the coupling effect of physiological frequencies were calculated. The amplitude-frequency characteristics are the amplitude values ​​at different frequencies, generating catheter displacement spectrum feature data. The instantaneous offset vector sequence at the tip, the real-time deflection angle sequence, and the duct displacement spectrum feature data of the same time period are synchronously fused according to timestamps. The fusion adopts a data splicing algorithm to associate the corresponding timestamp data of the three sequences. Multi-parameter correlation analysis is performed on the magnitude, directional persistence, deflection angle stability, and abnormal concentration of spectral energy of the offset vector. The offset vector magnitude is |Δr|=. The calculations are as follows: directional persistence is determined by whether the angle between the directional vectors at five consecutive sampling times is less than 10° (exemplary parameter, adjustable within the range of 8°-12°); the stability of the deflection angle is obtained by calculating the variance of the deflection angle at consecutive sampling times; and the abnormal concentration of spectral energy is determined by whether the power spectral density of a certain frequency is greater than twice the average power spectral density of all frequencies (exemplary parameter, adjustable within the range of 1.8-2.2 times). The integrated analysis results generate a displacement risk quantification dataset.

[0028] Furthermore, the calculation of the tip instantaneous offset vector sequence for each temporally phased trajectory segment includes: Extract the three-dimensional coordinate sequence of the catheter tip within a single respiratory cycle, and combine it with the ideal tip target coordinates and corresponding physiological range of motion defined in the baseline catheter-airway relative position relationship map to calculate the distance scalar value and direction vector between the real-time coordinate point and the ideal tip target coordinates. Determine whether the real-time coordinate points exceed the allowable range of physiological movement, mark abnormal offset points and analyze their direction vectors, and screen out high-risk directional offset points; By combining distance and direction information, the offset status corresponding to the high-risk directional offset point is quantitatively scored, and a tip instantaneous offset vector sequence is generated.

[0029] In this embodiment of the invention, a dynamic coordinate sequence of the catheter tip is generated by extracting the three-dimensional coordinate sequence (exemplary parameters, adjustable according to the sampling frequency) of all 300 sampling moments within a single respiratory cycle. Each coordinate sequence contains (xi, yi, zi), i = 1 to 300. The ideal tip target coordinates (x0, y0, z0) defined within the spatial relationship of the catheter-airway are extracted from the baseline catheter-airway relative positional relationship map. At the same time, the upper and lower limits of the physiological motion range allowed for the ideal tip target coordinates under the condition that the respiratory phase and cardiac phase are consistent with the current temporal trajectory segment are obtained from the physiological motion envelope boundary. When the respiratory phase is at the end of expiration, the upper and lower limits in the x direction are x0 ± 0.8 mm, the y direction is y0 ± 0.8 mm, and the z direction is z0 ± 0.8 mm (exemplary parameters, adjustable within the range of ±0.7-±0.9 mm). When the cardiac phase is during systole, the upper and lower limits in each direction are increased to ±1.0 mm (exemplary parameters, adjustable within the range of ±0.9-±1.1 mm). The dynamic boundary of the ideal coordinates of the current phase is generated. For each real-time coordinate point in the dynamic coordinate sequence of the tip, a sequence of 300 distance scalar values ​​is obtained by calculating its Euclidean distance to the ideal tip target coordinates (exemplary parameter, which can be adjusted according to the number of sampling times). Simultaneously, the spatial direction vector pointing from the ideal tip target coordinates to the real-time coordinate point is calculated. The direction vector calculation formula is u=(xi-x0,yi-y0,zi-z0), which, after normalization, yields u0=u / |u|, resulting in a sequence of direction vectors and generating a preliminary distance and direction dataset. Each distance scalar value in the preliminary distance and direction dataset is compared with the current phase ideal coordinate dynamic boundary. The difference between the distance scalar value and the corresponding upper and lower limits of the direction is calculated. If the distance scalar value is greater than the corresponding upper limit of the direction, it is marked as an abnormal offset point. The sampling time and coordinates of this point are recorded, generating abnormal offset point identification data. Each phased trajectory segment contains an average of 15 abnormal offset points (exemplary parameter, which can be adjusted according to the actual offset situation). For each abnormal offset point marked in the abnormal offset point identification data, its direction vector u0 is further analyzed. The anatomical danger direction is defined as the direction towards the contralateral bronchus (x direction less than x0), the airway wall (y direction greater than y0+1 mm), and below the carina (z direction greater than z0+1 mm) (exemplary parameters, which can be fine-tuned according to the airway anatomy). If the direction vectors of abnormal offset points at three consecutive sampling times are all pointing towards the same anatomical danger direction, and the angle between the direction vectors is less than 5° (exemplary parameters, which can be adjusted within the range of 4°-6°), then it is marked as a high-risk direction offset point, and a list of high-risk direction offset points is generated. Each temporal trajectory segment contains an average of 5 high-risk direction offset points (exemplary parameters, which can be adjusted according to the number of abnormal offset points).By combining the abnormal offset point identification data and the list of high-risk directional offset points, the offset status at each sampling moment in the dynamic coordinate sequence of the tip is quantitatively scored. The scoring formula is S=k1×d + k2×f, where k1 is the distance weight of 0.6, k2 is the directional risk weight of 0.4 (exemplary weights, which can be fine-tuned within ±0.05), d is the distance scalar value (unit: mm), and f is the directional risk coefficient. For non-abnormal offset points, f=0, for abnormal offset points, f=1, and for high-risk directional offset points, f=2. The scoring range is 0 to 5 points. Combining distance and direction information, a tip instantaneous offset vector sequence containing the offset severity and risk direction at each moment is generated. Each vector contains the offset magnitude, direction, and corresponding score.

[0030] Furthermore, the screening of high-risk directional offset points includes: Extract the direction vector of the abnormal offset point and calculate the unit direction vector, and compare it with the predefined set of anatomical danger direction vectors; Calculate the cosine of the spatial angle between the abnormal offset unit direction vector and each predefined danger direction vector, and determine the closest danger direction and similarity for each abnormal offset point; Set a similarity threshold for directional risk, mark abnormal offset points with similarity exceeding the threshold as high-risk directional offset points, and generate a list of high-risk directional offset points.

[0031] In this embodiment of the invention, the direction vector u=(Δx,Δy,Δz) of each abnormal offset point is extracted, where Δx=xi-x0, Δy=yi-y0, and Δz=zi-z0, xi,yi,zi are the real-time coordinates of the abnormal offset point, and x0,y0,z0 are the coordinates of the ideal tip target. The corresponding unit direction vector is calculated using the unit vector calculation formula u0=u / |u|, where |u|= This yields the abnormal offset unit direction vector u0=(u0x,u0y,u0z), where each abnormal offset point corresponds to a unit direction vector, generating a set of abnormal offset unit direction vectors. Four predefined anatomical danger direction vectors (exemplary parameters, 3-5 danger direction vectors can be set) are retrieved from the baseline duct-airway relative position relationship atlas. These vectors correspond to the duct tip moving deep into the bronchus, retracting into the main trachea, accidentally entering the contralateral bronchus, and pressing against the bronchial wall, respectively. The four danger direction vectors are v1=(0,0,1), v2=(0,0,-1), v3=(-1,0,0), and v4=(0,1,0). These four vectors are then organized into an anatomical danger direction vector set. For each abnormal offset unit direction vector u0, the cosine value of the spatial angle between u0 and each predefined danger direction vector vj in the dissection danger direction vector set is calculated using the formula cosθ=(u0·vj) / (|u0|×|vj|), where u0·vj=u0x×vjx + u0y×vjy + u0z×vjz, |u0|=1, |vj|=1. Therefore, the simplified formula is cosθ=u0x×vjx + u0y×vjy + u0z×vjz. Each abnormal offset point corresponds to 4 cosine values, generating an abnormal offset direction similarity dataset. In the abnormal offset direction similarity dataset, the four cosine values ​​of each abnormal offset point are compared, and the maximum value is selected. The predefined danger direction vector corresponding to this maximum value is the closest danger direction to the abnormal offset point. The maximum value and the corresponding danger direction are recorded to generate the closest danger direction and similarity data for each abnormal offset point. The cosine value ranges from -1 to 1, with values ​​closer to 1 indicating a more consistent direction. A direction risk similarity threshold of 0.8 is set (an example parameter, adjustable within the range of 0.75-0.85). The similarity of the closest danger direction to each abnormal offset point is compared with this threshold one by one, and the difference between the similarity and the threshold, Δcos=cosθ-0.8, is calculated. If Δcos≥0, i.e., the similarity exceeds the threshold, the offset direction of the abnormal offset point is determined to be highly close to a certain anatomical danger direction, and it is marked as a high-risk direction offset point. The sampling time, coordinates, closest danger direction, and similarity value of this point are recorded to generate a list of high-risk direction offset points. For abnormal offset points whose similarity does not exceed the threshold, they are determined to be ordinary abnormal offset points, and only their offset information is recorded; they are not included in the high-risk list. Meanwhile, the list of high-risk directional offset points is deduplicated, removing consecutively marked points of the same offset state due to excessively high sampling frequency. The deduplication criteria are: for high-risk offset points with three or more consecutive sampling times, if their coordinate deviation is less than 0.05 mm (exemplary parameter, adjustable within the range of 0.04-0.06 mm) and the included angle of the direction vector is less than 3° (exemplary parameter, adjustable within the range of 2°-4°), then only the marked point at the first sampling time is retained to avoid repeated counting leading to misjudgment of risk.

[0032] Furthermore, the intelligent decision-making and early warning module includes the following functions: The system receives a displacement risk quantification dataset and performs weighted fusion analysis on real-time offset vectors, real-time deflection angles, and duct dynamic displacement parameters to calculate the instantaneous displacement risk index and cumulative displacement risk trend at the current moment. The weighted fusion analysis is implemented using a neural network with a multi-layered analysis structure. This neural network includes a feature extraction layer, a temporal correlation analysis layer, and a risk index calculation layer connected sequentially. The temporal correlation analysis layer contains a temporal convolutional network module, which includes a causal convolutional layer, a dilated convolutional layer, and a residual connection layer connected sequentially. The dilation coefficient of the dilated convolutional layer increases exponentially with the number of layers. Based on the numerical range of the instantaneous displacement risk index value and the slope and shape of the cumulative displacement risk trend line, a preset multi-level risk judgment logic tree is triggered to map the continuous instantaneous displacement risk index value into discrete risk level classifications, and simultaneously output the specific anatomical direction of the displacement and the estimated displacement based on the motion trajectory prediction, and finally integrate to generate dynamic risk decision instructions. Based on the risk level, a preset alarm intensity configuration matrix is ​​matched, driving each alarm device to generate a multimodal alarm signal corresponding to the risk level; The multimodal alarm signals are formatted to generate differentiated correction prompts containing patient identification, alarm timestamp, risk level, displacement direction, real-time abnormal coordinates, and suggested adjustment vectors. These prompts are then pushed to the central monitoring system and anesthesiologist workstation, and corresponding alarm log files are generated.

[0033] In this embodiment of the invention, a displacement risk quantification dataset is received, and a weighted fusion analysis is performed on the real-time offset vector, real-time deflection angle, and duct dynamic displacement parameters. This weighted fusion analysis is implemented through a neural network containing a multi-layered analysis structure. The neural network includes a feature extraction layer, a time correlation analysis layer, and a risk index calculation layer connected in sequence. The feature extraction layer extracts the core features of each parameter and removes redundant information. The time correlation analysis layer includes a time convolutional network module, which comprises a causal convolutional layer, a dilated convolutional layer, and a residual connection layer connected in sequence. The dilation coefficient of the dilated convolutional layer increases exponentially with the number of layers (e.g., the dilation coefficient is 1 for the first layer, 2 for the second, and 4 for the third; the dilation coefficient can be adjusted according to the time-series correlation requirements), used to capture the long-term time-series correlation features of the risk parameters. The risk index calculation layer calculates the instantaneous displacement risk index (index range 0-10 points) based on the extracted features and time-series correlation information. Simultaneously, the time series of the instantaneous displacement risk index is integrated and trend-fitted to deduce the cumulative displacement risk trend, obtaining the cumulative risk value and the slope of the trend line. Based on the numerical range of the instantaneous displacement risk index value (0-3 points for low risk, 3-6 points for medium risk, and 6-10 points for high risk; the range can be adjusted according to clinical needs) and the slope and shape of the cumulative displacement risk trend line, a preset multi-level risk judgment logic tree is triggered. Through logic tree analysis, continuous instantaneous displacement risk index values ​​are mapped to discrete risk level classifications. At the same time, combined with the information of high-risk directional offset points, the specific anatomical direction of displacement is output (such as "mistakenly entering the contralateral bronchus" or "compressing the airway wall"). Based on the historical data of the catheter movement trajectory, the catheter offset within the next 10 seconds is predicted as the estimated displacement. By integrating information such as risk level, displacement direction, estimated displacement, risk development trend, and timestamp, dynamic risk decision instructions are generated. Based on the risk level and a preset alarm intensity configuration matrix, each alarm device is driven to generate a multimodal alarm signal corresponding to the risk level: at low risk, only the central monitoring system displays a green slight warning, and the audible, visual, and tactile alarms are not activated; at medium risk, the central monitoring system pops up a yellow pop-up window, the audible and visual alarm emits a medium-frequency warning tone (2 times / second, frequency adjustable within the range of 1.5-2.5 times / second), and the tactile feedback device vibrates for 3 seconds (vibration time adjustable within the range of 2-4 seconds); at high risk, all alarm devices are activated simultaneously, the audible and visual alarm emits a high-frequency warning tone (4 times / second, frequency adjustable within the range of 3.5-4.5 times / second), the red indicator light flashes rapidly, the tactile feedback device vibrates continuously, the central monitoring system pops up a red pop-up window and plays a voice prompt, and simultaneously links to the hospital nursing call system.The multimodal alarm signals are formatted to generate differentiated correction prompts. The messages include patient identification, alarm timestamp, risk level, displacement direction, real-time abnormal coordinates, and suggested adjustment vector (the adjustment vector is calculated as Δr'=-k×Δr, where k is the adjustment coefficient, which is 0.5 for low risk, 0.8 for medium risk, and 1.0 for high risk; the coefficient can be adjusted according to clinical intervention needs). The differentiated correction prompts are pushed to the central monitoring system and anesthesiologist workstations to ensure that medical staff receive alarm information in a timely manner. At the same time, corresponding alarm log files are generated and stored in JSON format, containing risk parameters, physiological signal fragments, intervention suggestions, etc., throughout the alarm process. These logs are pushed to the hospital's central monitoring system database for backup (the retention period is no less than 3 years, which can be adjusted according to the hospital's data management requirements) to support subsequent traceability and algorithm optimization.

[0034] Specifically, generating dynamic risk decision instructions involves extracting the instantaneous tip offset vector sequence, real-time deflection angle sequence, and catheter displacement spectral feature data from the displacement risk quantification dataset as input data. The amount of vector data in the instantaneous tip offset vector sequence can be adjusted according to the sampling frequency; in this embodiment, it exemplarily includes 300 vector data points. The amount of angle data in the real-time deflection angle sequence is consistent with the vector sequence; in this embodiment, it exemplarily includes 300 angle data points. The number of power spectral density values ​​in the catheter displacement spectral feature data can be adjusted according to the spectral analysis range; in this embodiment, it exemplarily includes power spectral density values ​​corresponding to 50 frequencies. A weighted fusion analysis neural network with a multi-layered analysis structure is constructed. The first layer is a feature extraction layer, the second layer is a time correlation analysis layer, and the third layer is a risk index calculation layer. The number of neurons in each layer of the network can be adjusted according to the feature extraction requirements; in this embodiment, they are exemplarily set to 64, 32, and 1, respectively. The three sets of input data are independently standardized through a feature extraction layer. The standardization formula is x'=(x-μ) / σ, where x is the original data, μ is the mean of the data set, and σ is the standard deviation of the data set. After standardization, a feature encoding algorithm is used to transform spatial offset, angular deviation, and spectral energy values ​​into feature vectors of a unified dimension (the dimension can be adjusted according to the number of neurons; in this embodiment, 64 dimensions are used as an example), ensuring that different types of data can be fused and analyzed. A temporal convolutional network module is introduced through a temporal correlation analysis layer. The time window length can be adjusted within the range of 8-12 sampling times; in this embodiment, it is set to 10 sampling times with a sliding step of 1. The encoded feature vectors are subjected to sliding analysis within continuous time windows to calculate the correlation of feature vectors within each window, capturing the temporal correlation pattern between instantaneous risk fluctuations and long-term risk trends, and generating temporally enhanced risk features (the dimension can be adjusted according to the number of neurons in the temporal correlation analysis layer; in this embodiment, 32 dimensions are used as an example).The risk index calculation layer performs a weighted fusion operation, setting dynamic weight coefficients (which can be adjusted according to clinical risk priority). In this embodiment, the weights are set as follows: the magnitude weight of the offset vector is 0.3, the directional persistence weight is 0.2, the stability weight of the deflection angle is 0.2, and the concentration weight of spectral energy in non-physiological frequency bands is 0.3. Non-physiological frequency bands can be defined according to physiological signal characteristics. In this embodiment, they are defined as below 0.1Hz and above 10Hz. The risk index calculation formula is RI=0.3×|Δr| + 0.2×D + 0.2×S +0.3×E, where |Δr| is the magnitude of the offset vector, D is the directional persistence coefficient (D=1 if the direction is consistent for 5 consecutive moments, otherwise D=0.5, and the number of consecutive moments can be adjusted), S is the deflection angle stability coefficient (the smaller the variance, the closer S is to 1), and E is the spectral energy concentration coefficient (the larger the proportion of power spectral density in non-physiological frequency bands, the closer E is to 1). The instantaneous displacement risk index value at the current moment is calculated and generated, with an index range of 0 to 10. The rate of change and direction of the instantaneous displacement risk index value within a preset time range (the preset time can be adjusted within the range of 8-12 seconds, and 10 seconds is used as an example in this embodiment) are integrated. The integral calculation formula is ∫RI(t)dt (t ranges from 0 to 10 seconds). Combined with historical risk data (the duration of historical data can be adjusted within the range of 3-7 minutes, and 5 minutes is used as an example in this embodiment), a linear fitting algorithm is used to predict the future risk trend (the prediction duration can be adjusted within the range of 0.5-1.5 minutes, and 1 minute is used as an example in this embodiment), generating a cumulative displacement risk trend line that characterizes the speed of risk development and the degree of potential harm. Based on the numerical range of the instantaneous displacement risk index (which can be adjusted according to clinical needs; in this embodiment, 0-3 points are exemplarily used for low risk, 3-7 points for medium risk, and 7-10 points for high risk) and the slope and shape of the cumulative displacement risk trend line (the slope can be adjusted within the range of 0.08-0.12; in this embodiment, greater than 0.1 indicates increased risk, and less than 0 indicates decreased risk), a preset multi-level risk judgment logic tree is triggered. This maps continuous quantitative risk data into discrete risk level classifications. Simultaneously, the specific anatomical direction of the displacement is determined through direction vector comparison. The future estimated displacement is predicted through a trajectory fitting algorithm (the prediction duration can be adjusted within the range of 8-12 seconds; in this embodiment, 10 seconds is exemplarily used). The formula for calculating the estimated displacement is Δs=v×t, where v is the past average displacement velocity (the average velocity calculation duration can be adjusted within the range of 3-7 seconds; in this embodiment, 5 seconds is exemplarily used), and t is the prediction duration. Finally, the risk level, displacement direction, and estimated displacement are integrated to generate a dynamic risk decision instruction.

[0035] The generation of differentiated correction prompts involves receiving dynamic risk decision instructions and using a data parsing algorithm to extract risk level, displacement direction, and estimated displacement data. Risk levels are categorized as low, medium, and high. Displacement directions correspond to four anatomically hazardous directions (adjustable based on a preset number of hazardous directions). The estimated displacement range is adjustable from 0.1 to 5 millimeters (the specific range is set according to clinical displacement risk assessment standards). Based on the risk level data, a preset alarm intensity configuration matrix is ​​matched: low risk corresponds to Level 1 alarm, medium risk to Level 2 alarm, and high risk to Level 3 alarm. Multimodal alarm parameter execution instructions are generated, including alarm level, audible and visual parameters, tactile parameters, and message format parameters. These multimodal alarm parameter execution instructions are sent in parallel to the audible and visual alarm driver unit, the tactile feedback device driver unit, and the digital alarm interface driver unit of the central monitoring system, ensuring synchronous response from all three units. The audible and visual alarm drive unit adjusts the alarm's operating status according to instructions. A level one alarm is characterized by flashing green lights at a frequency adjustable from 0.8 to 1.2 times per second (1 time per second in this embodiment), accompanied by a low-frequency alert sound with a volume adjustable from 45 to 55 decibels (50 decibels in this embodiment). A level two alarm is characterized by flashing yellow lights at a frequency adjustable from 1.8 to 2.2 times per second (2 times per second in this embodiment), accompanied by a mid-frequency alert sound with a volume adjustable from 65 to 75 decibels (70 decibels in this embodiment). A level three alarm is characterized by flashing red lights at a frequency adjustable from 3.8 to 4.2 times per second (4 times per second in this embodiment), accompanied by a high-frequency alarm sound with a volume adjustable from 85 to 95 decibels (90 decibels in this embodiment), generating a composite audible and visual alarm signal corresponding to the risk level. The tactile feedback device driving unit drives the device to generate tactile vibrations of specific intensity and pattern according to instructions. The vibration intensity of the first-level alarm can be adjusted within the range of 0.08-0.12g, and 0.1g is used in this embodiment for example. The frequency can be adjusted within the range of 8-12Hz, and 10Hz is used in this embodiment for example. The vibration intensity of the second-level alarm can be adjusted within the range of 0.28-0.32g, and 0.3g is used in this embodiment for example. The frequency can be adjusted within the range of 18-22Hz, and 20Hz is used in this embodiment for example. The vibration intensity of the third-level alarm can be adjusted within the range of 0.48-0.52g, and 0.5g is used in this embodiment for example. The frequency can be adjusted within the range of 38-42Hz, and 40Hz is used in this embodiment for example. This forms a physical alarm that is synchronized with the sound and light alarm and can be perceived by touch.The central monitoring system's digital alarm interface driver unit generates standard alarm messages in a formatted manner according to instructions. The message includes the current patient identifier (consisting of 8 digits, the number of digits can be adjusted according to hospital patient management standards), alarm timestamp (accurate to milliseconds), specific risk level, displacement direction code (which can be set according to the number of dangerous directions; in this embodiment, 01 is exemplarily used for deep bronchus, 02 for main bronchus retraction, 03 for contralateral bronchus, and 04 for bronchial wall pressure), real-time abnormal coordinates, and a system-generated suggested adjustment vector. The suggested adjustment vector is calculated using the formula Δr'=-k×Δr, where k is the adjustment coefficient: low risk k=0.5, medium risk k=0.8, and high risk k=1.0 (the coefficient can be adjusted according to clinical intervention needs). The standard alarm message is pushed to the central monitoring system's display terminal and historical log database, and simultaneously forwarded to the anesthesiologist's workstation. A structured alarm log file is generated and stored using a data formatting algorithm. The log file contains all alarm parameters and time records for easy subsequent traceability.

[0036] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A smart monitoring and alarm system for the displacement of a dual-lumen bronchial tube, characterized in that, Includes the following modules: The catheter positioning information acquisition module is used to dynamically track the three-dimensional spatial coordinates of the double-lumen bronchial catheter in the insertion state using a three-dimensional spatial positioning unit, collect the real-time position fluctuation signals of the catheter tip and body during the physiological cycle, and generate the original catheter spatial trajectory dataset. The morphological baseline model construction module is used to construct an individualized airway-catheter matching morphological simulation model for the patient based on the patient's preoperative 3D thoracic cavity image and airway anatomical parameters, and generate a baseline catheter-airway relative positional relationship map. The displacement risk calculation engine module is used to input the original catheter spatial trajectory dataset into the baseline catheter-airway relative position relationship map for spatiotemporal registration, calculate the real-time offset vector of the catheter tip relative to the target bronchial opening, the real-time deflection angle of the catheter body relative to the central axis of the main trachea, and the dynamic displacement parameters of the catheter under the physiological motion coupling effect, and generate a displacement risk quantification dataset. The intelligent decision-making and early warning module is used to receive the displacement risk quantification dataset, perform weighted fusion analysis on the real-time offset vector, real-time deflection angle and duct dynamic displacement parameters, calculate the instantaneous displacement risk index and cumulative displacement risk trend at the current moment, and generate dynamic risk decision instructions including risk level, displacement direction and estimated displacement. Upon receiving dynamic risk decision instructions, when the instantaneous displacement risk index exceeds the preset safety threshold or the cumulative displacement risk trend shows a dangerous slope, an alarm signal is output and differentiated correction prompt information is generated, and an alarm log file containing abnormal coordinates and suggested adjustment vectors is sent.

2. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 1, characterized in that, The three-dimensional spatial positioning unit adopts a distributed optical fiber sensor network, which is deployed around the outer wall of the patient's trachea and arranged in a ring array.

3. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 2, characterized in that, The catheter positioning information acquisition module includes the following functions: The distributed optical fiber sensing network is based on a preset physiological curvature and fits onto the patient's corresponding neck and upper chest tracheal projection area. It emits detection light and receives backscattered signals formed by reflections from deep tissues and the duct wall. The backscattered signals are processed by phase demodulation and wavefront reconstruction to separate the strong reflection signal components generated by the corresponding feature markers of the double-lumen bronchial duct, thus obtaining the duct feature reflection signal sequence. The time and frequency domains of the catheter characteristic reflection signal sequence are jointly analyzed to extract the characteristic parameters corresponding to each signal component. The preliminary coordinates of each reflection point in three-dimensional space are calculated by the optical domain reflection algorithm to generate a set of spatial coordinates of catheter scatter points. Based on the temporal continuity of the catheter scatter spatial coordinate set, the trajectory of the key marker points in the catheter tip region and the catheter body is fitted to generate a continuous spatial curve describing the catheter morphology. The continuous spatial curve is sampled at high frequency within a single respiratory cycle. The coordinate changes of the catheter tip and multiple preset main marker points in the three-dimensional coordinate system over time are recorded. The physiological phase marker signal of this period is collected synchronously as a time synchronization reference to generate an original catheter spatial trajectory dataset with physiological phase markers. The sampling frequency of the high-frequency sampling is not less than 1 kHz.

4. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 1, characterized in that, The morphological benchmark model construction module includes the following functions: Import the patient's preoperative thin-section chest computed tomography (DICOM) image data and extract the three-dimensional airway cavity model from the glottis to the bilateral main bronchi and lobar bronchi to generate an individualized three-dimensional airway geometric model. In the individualized three-dimensional airway geometric model, according to the model and specification parameters of the double-lumen bronchial tube, call the catheter three-dimensional entity database to match the corresponding model of the catheter three-dimensional entity model. Based on the expected insertion depth and lateral information of the catheter in the anesthesia record, a virtual insertion operation is performed in the individualized three-dimensional geometric model of the airway. The three-dimensional solid model of the catheter is inserted into the target main bronchus according to the clinical operation specifications, and the initial pose data of the virtual catheter is generated. Based on the initial pose data of the virtual catheter, the spatial relationship between the catheter and the airway is simulated under ideal alignment, and the normal physiological motion envelope boundary of the catheter is defined. The spatial relationship between the catheter and the airway and the normal physiological motion envelope boundary of the catheter are integrated to generate a baseline relative positional relationship map of the catheter and the airway.

5. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 4, characterized in that, The simulated catheter-airway spatial relationship under ideal alignment includes: From the initial pose data of the virtual catheter and the individualized three-dimensional geometric model of the airway, discrete point cloud datasets representing the geometric features of the model surface are extracted respectively. Based on the discrete point cloud datasets, the Euclidean distance from each discrete point on the catheter surface to all discrete points on the inner wall of the airway is calculated. The minimum distance value corresponding to each point on the catheter surface and its corresponding nearest point coordinates on the inner wall of the airway are selected to generate a preliminary point-to-point distance mapping table. The initial point-to-point distance mapping table is filtered to remove non-contact abnormal long-distance point pairs, and point pairs that are suspected to represent potential contact or proximity relationships are retained to generate a dataset of effective contact and proximity point pairs. Based on the effective contact proximity point pair dataset, the direction vectors of the catheter's long axis and the airway's main axis are obtained, and the three-dimensional spatial angle between them is calculated. The projection components of this angle on the coronal and sagittal planes are also calculated to generate the spatial angle data of the catheter-airway axis. The distance statistics parameters of all point pairs in the effective contact proximity point pair dataset are integrated with the spatial angle data of the catheter-airway axis to generate the spatial relationship of the catheter-airway under ideal alignment.

6. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 5, characterized in that, The filtering process for the initial point-to-point distance mapping table includes: Obtain the distance values ​​recorded in the preliminary point-to-point distance mapping table, draw a histogram of the statistical distribution of all distance values, and set a dynamic filtering threshold based on the distance percentile according to the distribution characteristics; compare the distance values ​​between each pair of points with the dynamic filtering threshold, and filter out a list of abnormal point pairs to be removed; For each point pair in the list of abnormal point pairs to be removed, spatial position relationship analysis is performed, invalid neighboring point pairs are removed, and point pairs near the expected contact area of ​​the catheter are retained and subjected to secondary discrimination. By integrating the valid points after secondary discrimination, a dataset of valid contact neighbor pairs is generated.

7. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 1, characterized in that, The displacement risk calculation engine module includes the following functions: Receive the raw catheter spatial trajectory dataset with physiological phase markers, and call the catheter-airway spatial relationship and the normal physiological motion envelope boundary of the catheter in the baseline catheter-airway relative position relationship map as the registration and calculation benchmark; The original ductal spatial trajectory dataset is temporally segmented to generate a temporally phased trajectory fragment dataset. For each temporal trajectory segment, calculate the instantaneous offset vector sequence of the catheter tip's real-time coordinates and the ideal tip target coordinates, as well as the real-time deflection angle sequence of the catheter body; Spectral analysis of the catheter tip's motion trajectory is performed to extract characteristic parameters of the trajectory signal at physiological frequencies and their harmonic frequencies, thereby generating dynamic displacement parameters of the catheter. The instantaneous offset vector sequence of the tip, the real-time deflection angle sequence, and the dynamic displacement parameters of the catheter are synchronously fused according to the timestamp, and multi-parameter correlation analysis is performed to generate a displacement risk quantification dataset.

8. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 7, characterized in that, The calculation of the tip instantaneous offset vector sequence for each temporalized trajectory segment includes: Extract the three-dimensional coordinate sequence of the catheter tip within a single respiratory cycle, and combine it with the ideal tip target coordinates and corresponding physiological range of motion defined in the baseline catheter-airway relative position relationship map to calculate the distance scalar value and direction vector between the real-time coordinate point and the ideal tip target coordinates. Determine whether the real-time coordinate points exceed the allowable range of physiological movement, mark abnormal offset points and analyze their direction vectors, and screen out high-risk directional offset points; By combining distance and direction information, the offset status corresponding to the high-risk directional offset point is quantitatively scored, and a tip instantaneous offset vector sequence is generated.

9. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 8, characterized in that, The high-risk directional deviation points identified include: Extract the direction vector of the abnormal offset point and calculate the unit direction vector, and compare it with the predefined set of anatomical danger direction vectors; Calculate the cosine of the spatial angle between the abnormal offset unit direction vector and each predefined danger direction vector, and determine the closest danger direction and similarity for each abnormal offset point; Set a similarity threshold for directional risk, mark abnormal offset points with similarity exceeding the threshold as high-risk directional offset points, and generate a list of high-risk directional offset points.

10. The intelligent monitoring and alarm system for dual-lumen bronchial tube displacement according to claim 1, characterized in that, The intelligent decision-making and early warning module Includes the following features: The system receives a displacement risk quantification dataset and performs weighted fusion analysis on real-time offset vectors, real-time deflection angles, and duct dynamic displacement parameters to calculate the instantaneous displacement risk index and cumulative displacement risk trend at the current moment. The weighted fusion analysis is implemented using a neural network with a multi-layered analysis structure. This neural network includes a feature extraction layer, a temporal correlation analysis layer, and a risk index calculation layer connected sequentially. The temporal correlation analysis layer contains a temporal convolutional network module, which includes a causal convolutional layer, a dilated convolutional layer, and a residual connection layer connected sequentially. The dilation coefficient of the dilated convolutional layer increases exponentially with the number of layers. Based on the numerical range of the instantaneous displacement risk index value and the slope and shape of the cumulative displacement risk trend line, a preset multi-level risk judgment logic tree is triggered to map the continuous instantaneous displacement risk index value into discrete risk level classifications, and simultaneously output the specific anatomical direction of the displacement and the estimated displacement based on the motion trajectory prediction, and finally integrate to generate dynamic risk decision instructions. Based on the risk level, a preset alarm intensity configuration matrix is ​​matched, driving each alarm device to generate a multimodal alarm signal corresponding to the risk level; The multimodal alarm signals are formatted to generate differentiated correction prompts containing patient identification, alarm timestamp, risk level, displacement direction, real-time abnormal coordinates, and suggested adjustment vectors. These prompts are then pushed to the central monitoring system and anesthesiologist workstation, and corresponding alarm log files are generated.

Images