Endoscope catheter calibration and drift compensation methods and related devices

Abstract

The embodiment of the application discloses an endoscope catheter calibration and drift compensation method and related device. Through multi-source sensing observation data, pose estimation and drift compensation output are completed in a fast ring period, drift discrimination and incremental update of calibration parameters are completed in a slow ring period, online continuous calibration of the calibration parameters is realized, accumulated drift caused by various factors in the operation process can be offset in real time, the pose estimation accuracy of the endoscope catheter is maintained throughout the operation, and gradual invalidation of the parameters of single preoperative calibration with operation advancement is avoided. Meanwhile, short window residual features and long window residual features are set to respectively discriminate instantaneous abnormalities and chronic drift, parameter updating is triggered only when it is confirmed that there is chronic drift and there is no obvious instantaneous abnormality, false parameter adjustment under abnormal working conditions is avoided, and the pose output is further ensured to be smooth in the parameter updating process through a gate smoothing writing mechanism, sudden jump interference in the operation is avoided, and the effectiveness of the drift compensation and the safety of the operation are considered.

Description

Technical Field

[0001] This application relates to the field of medical devices, specifically to a method and apparatus for calibrating and compensating for endoscopic catheter drift. Background Technology

[0002] A bronchoscopic robotic system typically includes preoperative CT modeling, intraoperative navigation, sensor feedback, and execution control. The bronchial catheter, a slender, flexible tube inserted into the patient's bronchus within a flexible bronchoscope, generally integrates actuation, sensing, and working channels. It is the core component directly involved in lesion exploration and sampling, and the accuracy of its distal end's positioning directly determines the precision and safety of the surgical procedure.

[0003] During the long-path advancement, retraction, and multiple bends of the catheter, the actual position of the catheter tip is affected by factors such as friction, tissue constraint, rebound, and respiratory motion, thus deviating from the ideal kinematic prediction. If this deviation cannot be continuously corrected, the accuracy of tip arrival, sampling stability, and safety will all decrease significantly.

[0004] When a catheter travels along a long path within the body, with multiple bends and complex tissue constraints, the original static calibration parameters will become invalid, and pose and control errors will accumulate continuously. If only a single sensor source or intermittent correction at key nodes is relied upon, the system will struggle to identify chronic drift in a timely manner and will be unable to maintain continuous, stable, and safe closed-loop control under conditions of sensor conflict, respiratory disturbance, and signal lockout. Summary of the Invention

[0005] This application provides an endoscopic catheter calibration and drift compensation method and related device. By combining a fast and slow dual-cycle architecture with long and short window residual feature discrimination, online continuous calibration of drift is achieved, while ensuring the stability of pose output and the safety of surgical operation.

[0006] The first aspect of this application provides a method for calibrating and compensating for the drift of an endoscopic catheter, the method comprising:

[0007] In each fast loop cycle, acquire multi-source sensor observation data for the endoscopic catheter;

[0008] The pose estimation result of the endoscopic catheter is calculated based on the multi-source sensor observation data, and the pose estimation result is corrected and compensated to obtain the compensated pose estimation result.

[0009] Determine the predicted pose of the endoscopic catheter, and determine the pose residual of the fast loop cycle based on the predicted pose and the compensated pose estimation result;

[0010] After a preset number of fast loop cycles, a slow loop cycle is entered. In the slow loop cycle, short window residual features and long window residual features are statistically obtained based on the pose residuals of multiple fast loop cycles. The short window residual features are used for instantaneous anomaly detection, and the long window residual features are used for chronic drift detection.

[0011] If the long window residual feature is determined to exceed the drift threshold based on the chronic drift discrimination, and the number of short window residual abnormalities is determined not to exceed the preset abnormality number threshold based on the instantaneous anomaly discrimination, then the calibration parameters of the pose estimation are updated to obtain the updated calibration parameters.

[0012] The updated calibration parameters are gradually updated into the pose estimation calculation of the first fast cycle after the slow cycle through a gated smooth write mechanism.

[0013] A second aspect of this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method of the first aspect described above.

[0014] A third aspect of this application provides a computer storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described in the first aspect.

[0015] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0016] By utilizing multi-source sensor data, pose estimation and drift compensation output are performed in the fast loop cycle, while drift discrimination and incremental updates of calibration parameters are performed in the slow loop cycle. This enables continuous online calibration of calibration parameters, effectively offsetting cumulative drift caused by various factors during surgical advancement. It maintains the pose estimation accuracy of the endoscopic catheter throughout the procedure, preventing pre-operatively calibrated parameters from gradually becoming invalid as the surgery progresses, thus ensuring the accuracy and safety of interventional navigation surgery. Simultaneously, short-window and long-window residual features are used to distinguish between transient anomalies and chronic drift, respectively. Parameter updates are triggered only when chronic drift is confirmed and there are no obvious transient anomalies, avoiding erroneous parameter adjustments under abnormal conditions. Combined with a gated smooth writing mechanism, this further ensures stable pose output during parameter updates, preventing abrupt changes that could interfere with surgical operations, thus balancing the effectiveness of drift compensation with the safety of surgical procedures. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the endoscope catheter calibration and drift compensation method in the embodiments of this application;

[0018] Figure 2This is an exemplary timing flowchart of the endoscope catheter calibration and drift compensation method in the embodiments of this application;

[0019] Figure 3 This is a schematic diagram of the structure of a computer device in an embodiment of this application;

[0020] Figure 4 This is another schematic diagram of the structure of the computer device in the embodiments of this application. Detailed Implementation

[0021] This application provides an endoscopic catheter calibration and drift compensation method and related device. By combining a fast and slow dual-cycle architecture with long and short window residual feature discrimination, online continuous calibration of drift is achieved, while ensuring the stability of pose output and the safety of surgical operation.

[0022] Currently, the industry mainly offers three types of existing technical solutions for endoscopic catheter positioning correction:

[0023] Category 1 Existing Technology: One-time preoperative calibration plus open-loop control;

[0024] This type of approach aligns the coordinate system, calibrates sensor offsets, and sets control mapping parameters before surgery. During the procedure, it primarily follows the preoperative results and control commands, making only limited manual corrections when necessary. Its advantages include a simple system structure and low implementation threshold. However, this approach assumes that the preoperative calibration parameters remain valid during surgery and cannot be updated in real-time according to changing conditions. When the catheter undergoes multiple bends, changes in local stress states, or changes in tissue constraints within the body, the model's predicted pose gradually deviates from the actual pose, and the cumulative error increases significantly. This type of method struggles to meet the accuracy requirements of dynamic navigation throughout the entire process.

[0025] The second category of existing technologies: online estimation using a single sensor source;

[0026] This type of approach typically uses a single sensor as the primary observation source, such as electromagnetic positioning, shape sensing, or single visual feedback, and updates the state using Kalman filtering or other simple estimators. Its advantages include a relatively simple system and low cost. However, the robustness of this approach is significantly limited by its dependence on a single-source environment. When electromagnetic interference occurs, visual obstruction occurs, or local nonlinear deformation exceeds the assumptions of the single model, the estimation results become unstable or even distorted. Due to insufficient observation dimensions, the system struggles to obtain effective corrections from other information sources, easily leading to incorrect control direction and overly conservative operation.

[0027] The third type of existing technology: event-triggered intermittent correction;

[0028] This type of approach typically pauses the procedure at critical points, when confidence levels drop, or when the operator actively triggers it, performing a correction using fluoroscopy, ultrasound, or other imaging techniques before resuming navigation. Its advantage lies in its strong verification capability at critical points. However, the compensation in this approach is intermittent rather than continuous. Drift continues to accumulate between corrections, requiring frequent pauses for confirmation by the operator, leading to poor operational continuity, decreased surgical efficiency, and subjective variations in manual adjustments, making it difficult to establish a stable and consistent closed loop.

[0029] Therefore, addressing the aforementioned related technical solutions and their shortcomings, this application proposes an endoscopic navigation method. Specifically, to address the shortcomings of the first type of technology, namely, the assumption that the preoperative calibration parameters remain unchanged during the procedure, leading to a gradual deviation between the model's predicted pose and the actual pose after multiple catheter bends, advancements, and tissue contacts, ultimately resulting in increased cumulative error, worsened end-effector arrival deviation, and an increased number of repeated adjustments, this application addresses this by establishing an automatic online calibration mechanism during the procedure. This mechanism ensures that key sensing and control mapping parameters are updated and converged in real time according to changes in operating conditions, thereby suppressing cumulative errors during long-path operation.

[0030] To address the shortcomings of the second type of technology, namely that the system relies on only a single sensing source, leading to observational instability or bias when electromagnetic interference, visual obstruction, or deformation exceeds model assumptions, ultimately resulting in incorrect control correction direction, local oscillations, or ineffective drift control, this application's embodiments construct a multi-source fusion state estimation and multi-dimensional confidence weighting mechanism. This mechanism ensures that the signal quality, inter-source consistency, and temporal synchronization of each source jointly determine the fusion weight, thereby improving the robustness of drift identification and compensation.

[0031] To address the shortcomings of the third type of technology—namely, the lack of continuous error constraints between corrections due to event-triggered intermittent correction, which allows drift to accumulate continuously within the correction interval, ultimately leading to frequent pauses, reduced efficiency, and increased manual verification burden—this application's embodiments establish a closed loop of continuous drift monitoring and micro-compensation. Drift is identified through a combination of short-window residuals and long-window trends, enabling continuous small-scale compensation without interrupting the operation process, thus reducing reliance on interruptive corrections.

[0032] The endoscope catheter calibration and drift compensation method in the embodiments of this application is described below:

[0033] Please see Figure 1 One embodiment of the endoscopic catheter calibration and drift compensation method in this application includes:

[0034] 101. Acquire multi-source sensor observation data for the endoscopic catheter during each fast loop cycle;

[0035] In this embodiment, drift refers to the cumulative error caused by catheter material deformation, joint space wear, sensor element temperature drift, and human tissue interaction during interventional navigation surgery. As the surgery progresses, the calibration parameters will gradually deviate from the initial calibration value, ultimately leading to a decrease in the accuracy of the endoscopic catheter pose estimation, affecting navigation accuracy and surgical safety.

[0036] Among them, calibration parameters refer to the calibration parameters required to convert the raw sensor observation data into the actual position and orientation of the catheter. These mainly include the installation zero-position deviation of the shape sensing fiber, the bending curvature-deformation conversion coefficient, the spatial installation offset of each sensing segment, as well as the joint zero-position deviation of the encoder and the transmission clearance compensation coefficient. These parameters will slowly shift due to factors such as temperature changes during surgical advancement, repeated force deformation of the catheter, and slight loosening of the installation connectors, which will lead to cumulative drift in the position and orientation estimation results. Continuous compensation can be achieved through online updates.

[0037] The fast loop cycle refers to the basic cycle used for pose estimation and control. The duration of the fast loop cycle is preferably set between 50ms and 100ms, but other durations are also possible. It meets the real-time requirements of interventional surgical navigation, ensuring that the pose output matches the catheter advancement speed without significant control lag. Conversely, the slow loop cycle refers to the cycle used for online calibration and drift detection. A slow loop cycle is triggered only after a preset number of fast loop cycles. The duration of the slow loop cycle is typically set to 10 to 100 times the fast loop cycle duration (or other multiples), which allows for sufficient residual data to complete reliable drift detection without excessively consuming system computing resources.

[0038] An endoscopic catheter is a flexible surgical instrument used in interventional surgeries within human cavities. It can be inserted through natural body cavities or small incisions and, under the control of an external drive device, performs maneuvers such as advancement and turning. The tip of the catheter carries a working execution component to complete the target surgical procedure. The endoscopic catheter described in this application integrates shape-sensing fiber optics and joint rotation encoders in different sensing segments, and is compatible with other observation sources such as preoperative images and intraoperative external imaging, thereby achieving multi-source data fusion acquisition. For example, this endoscopic catheter can be a bronchial endoscope catheter.

[0039] Therefore, multi-source sensor observation data refers to the shape curvature data of each sampling point of the catheter output by the shape sensing fiber, and the joint rotation angle data output by the encoder. In the case of additional observation sources, it can also include the head position marker data detected by external imaging equipment during the operation. Specifically, it includes the three-dimensional coordinate data of the continuous shape points of the entire catheter segment acquired by the shape sensing fiber, the joint rotation angle data acquired by the encoder, and can also selectively introduce the pose observation data acquired by intraoperative X-ray, ultrasound, or electromagnetic positioning. All data are timestamped through a preset synchronous frame header identifier to ensure that all observation data within the same fast loop cycle correspond to the actual shape of the catheter at the same moment.

[0040] Computer equipment can acquire multi-source sensor observation data by directly receiving synchronized data that has been time-aligned from the corresponding sensor acquisition module through a real-time communication interface, or by actively polling each sensor device to trigger data acquisition for the current fast loop cycle and synchronously timestamping it. This application does not make any specific limitations on this.

[0041] 102. Calculate the pose estimation result of the endoscopic catheter based on the multi-source sensor observation data, and correct and compensate the pose estimation result to obtain the compensated pose estimation result;

[0042] The reason why the pose estimation result of the endoscopic catheter can be calculated based on multi-source sensor observation data is that the shape sensing fiber can directly acquire the three-dimensional shape of the entire catheter, thus directly resolving the pose estimation result from the tip position; the joint angles output by the encoder can be combined with the kinematic model to directly calculate the pose estimation result of the catheter, and different sensor sources provide pose observation information from different dimensions. After obtaining the initial pose observation results of each sensor source, the final pose estimation result is obtained by fusing the data according to the preset weighted fusion rules and the real-time confidence weights of each sensor source.

[0043] For example, different sensor sources independently output pose-related observation results. Consistent weighted fusion reduces the impact of errors from a single sensor source. The core principle is to dynamically adjust the weights based on the real-time operating status of the sensor sources, avoiding interference from abnormal observations on the final pose result. First, continuous shape point data acquired from the shape-sensing fiber can be directly reconstructed using morphological interpolation to obtain the three-dimensional morphology and tip pose of the entire catheter. This observation path does not rely on the catheter kinematics model and can directly respond to morphological changes caused by catheter deformation under stress. Then, the joint angles acquired by the encoder are substituted into the pre-trained catheter kinematics model to calculate an estimate of the overall catheter morphology and tip pose. If additional intraoperative imaging or electromagnetic positioning observation sources are available, an additional set of direct observations of the tip position can be obtained. By weighting and fusing all observation results according to the adjusted real-time weights, the fused initial pose estimation result can be obtained.

[0044] After initial pose estimation is completed, the pose estimation result is corrected and compensated based on the previously accumulated drift correction, resulting in the compensated pose estimation result output in the current fast loop cycle. The purpose of correction and compensation is to counteract the chronic drift effect caused by accumulated calibration parameter deviations, ensuring that the pose estimation result always matches the actual position of the current catheter, and preventing the accumulated error from increasing as the surgery progresses, ultimately leading to navigation failure. The principle is as follows: initial calibration parameters such as sensor installation zero position and catheter deformation coefficient are obtained through preoperative calibration. During the surgical process, the slow loop cycle continuously monitors the cumulative trend of pose residuals, and uses the residual characteristics to infer the deviation of the calibration parameters, obtaining the parameter increments that need to be updated. When performing pose calculation in each fast loop cycle, the initial estimation result is corrected based on the current calibration parameters, ultimately obtaining an accurate pose result that has been compensated for the drift effect.

[0045] Therefore, the drift correction amount refers to the cumulative correction component calculated from the deviation between the calibration parameters and the initial calibration parameters. It can reflect the systematic drift caused by factors such as sensor installation position offset and changes in catheter material properties during the operation. When superimposed on the initial posture estimation result, it can effectively offset such deviations.

[0046] Optionally, during the correction and compensation process, the correction results can be filtered and smoothed by combining the confidence of multi-source fusion, while limiting the magnitude of a single correction to avoid abrupt pose results affecting the doctor's operating feel and surgical judgment.

[0047] 103. Determine the predicted pose of the endoscopic catheter, and determine the pose residual of the fast loop cycle based on the predicted pose and the compensated pose estimation result;

[0048] The predicted pose refers to the theoretical pose result calculated based on the kinematic model. It is calculated by using the estimated pose after compensation in the previous cycle as a benchmark and combining it with the drive input of the catheter joint in the current cycle. It can reflect the position that the catheter should theoretically reach under the drive of control commands. When chronic drift exists, the calibration parameters will deviate, causing the actual fused compensated pose estimation result to gradually deviate from the predicted pose. The pose residual will gradually increase with the accumulation of drift. Therefore, the existence and degree of drift can be judged by the change characteristics of the pose residual.

[0049] In the specific calculation, the difference between the compensated pose estimation result and the predicted pose at the corresponding position can be directly calculated to obtain the position residual and the attitude angle residual respectively. The two are then fused according to the preset weights to obtain the final pose residual of the current fast cycle, which is used for subsequent drift feature statistics and discrimination.

[0050] 104. After a preset number of fast loop cycles, a slow loop cycle is entered. In the slow loop cycle, short window residual features and long window residual features are statistically obtained based on the pose residuals of multiple fast loop cycles. The short window residual features are used for instantaneous anomaly detection, and the long window residual features are used for chronic drift detection.

[0051] In this step, the pose residual data of the most recent N fast loop cycles and the most recent M fast loop cycles (M>N) are first updated by sliding according to the preset statistical window. The residual statistical features in the corresponding window are extracted respectively. That is, the pose residual of the most recent N fast loop cycles can be used to obtain the short window residual features, and the pose residual of the most recent M fast loop cycles can be used to obtain the long window residual features. The short window residual features capture the sudden fluctuation of the residual in a short period of time, which corresponds to instantaneous anomalies such as operation collision and sensor jump. The long window residual features capture the cumulative trend of the residual over a longer period of time, which corresponds to chronic drift such as sensor zero position offset and conduit material creep.

[0052] After feature extraction is completed, the transient anomaly discrimination rule is first applied to compare the short-window residual features against a threshold. If the transient anomaly discrimination condition is met, the number of anomalies within the current slow window is accumulated, and the current fast loop cycle is marked as an anomalous frame. If the threshold condition is not met, it is determined that there is no transient anomaly, and the subsequent chronic drift discrimination is continued. In the chronic drift discrimination, the long-window residual features are compared with the preset drift discrimination threshold. If the discrimination condition is met, it is confirmed that there is a chronic drift that needs to be corrected, and the subsequent calibration parameter update process is triggered. If the threshold condition is not met, the original calibration parameters are maintained, and the current slow loop cycle process is directly terminated.

[0053] 105. If the long window residual feature is determined to exceed the drift threshold based on the chronic drift discrimination, and the number of short window residual abnormalities is determined not to exceed the preset abnormality number threshold based on the instantaneous anomaly discrimination, then the calibration parameters of the pose estimation are updated to obtain the updated calibration parameters.

[0054] In this step, after confirming the existence of drift and that the transient anomaly is within the allowable range, the calibration parameter update process can be initiated to avoid incorrect parameter updates under abnormal conditions that could lead to greater deviations. For example, firstly, the direction and step size of the calibration parameters that need to be corrected are obtained by fitting the trend characteristics of the long-window residuals. Then, based on the calibration parameters currently in use, incremental adjustments are made according to the obtained correction direction and step size to correct the systematic deviation components caused by the accumulation of chronic drift.

[0055] Optionally, to avoid abrupt changes in pose results due to excessive parameter adjustments, a maximum allowable range can be set for the correction amount of the calibration parameters. This means the absolute value of the correction amount must not exceed a preset maximum correction threshold, ensuring that each parameter update is within a safe range and does not cause irreversible damage to the calibration results. If the calculated correction amount exceeds this maximum allowable range, the correction amount is clipped to the maximum allowable range, and the parameters are adjusted only by the maximum allowable magnitude, avoiding abrupt changes in pose caused by excessive single adjustments.

[0056] After updating the calibration parameters, the validity of the update is determined by verifying the residuals. If the verification finds that the pose residuals increase instead of decrease, the parameter update is terminated, and the calibration parameters before the update are retained to avoid amplifying the drift error by incorrect parameter updates.

[0057] 106. The updated calibration parameters are gradually updated into the pose estimation calculation of the first fast cycle after the slow cycle through a gated smooth write mechanism;

[0058] In this step, the original calibration parameters are not directly replaced in one step to complete the update. Instead, multiple fast-loop cycles are used to make gradual, small-scale corrections. This avoids sudden changes in pose estimation results caused by a one-time parameter replacement, which could affect the stability and safety of the surgical procedure. The gated smooth write mechanism sets the write step size for each fast-loop cycle based on the current chronic drift level. The higher the drift level, the larger the incremental proportion of a single write, but it still controls the adjustment range within a preset safety range. This ensures drift compensation efficiency while preventing sudden changes in pose results from interfering with the surgeon's operation. During the gradual write process, changes in pose residuals are monitored in real time during each fast-loop cycle. If the residuals show an abnormal increase, the update is immediately stopped, and the currently written parameter state is retained or the parameters are reverted to the state before the update.

[0059] Therefore, in this embodiment, pose estimation and drift compensation output are completed in the fast loop cycle using multi-source sensor observation data, while drift discrimination and incremental updates of calibration parameters are completed in the slow loop cycle. This achieves online continuous calibration of calibration parameters, which can offset the cumulative drift caused by various factors during surgical advancement in real time, maintain the pose estimation accuracy of the endoscopic catheter throughout the entire process, and prevent the parameters calibrated once before surgery from gradually becoming invalid as the surgery progresses, thus ensuring the accuracy and safety of interventional navigation surgery. At the same time, short-window residual features and long-window residual features are set to distinguish between transient anomalies and chronic drift, respectively. Parameter updates are only triggered when chronic drift is confirmed and there are no obvious transient anomalies, avoiding erroneous parameter adjustments under abnormal conditions. Combined with a gated smooth writing mechanism, this further ensures that the pose output is stable during parameter updates, without causing sudden jumps that interfere with surgical operations, thus balancing the effectiveness of drift compensation and the safety of surgical operations.

[0060] based on Figure 1In one optional implementation of the illustrated embodiment, the computer device may further acquire target location information, which represents the final target location to which the tip of the endoscopic catheter needs to reach, as determined by the preoperative planning path. This target location could be, for example, the location of the surgical lesion. After each fast-loop cycle of pose estimation calculation, based on the compensated pose estimation result and the target location information, it is determined whether the distance deviation between the current position of the tip of the endoscopic catheter and the target position is within a preset range. If yes, it is determined that the endoscopic catheter has reached the target position, and the compensated pose estimation result and arrival prompt information are output. If not, the pose estimation and pose compensation process for the next fast-loop cycle continues.

[0061] In other words, the deviation between the actual position of the endoscopic catheter tip and the planned target position can be calculated based on the current compensated pose estimation result. The actual position of the endoscopic catheter tip is directly obtained by parsing the pose estimation result after the current fast loop cycle compensation. The actual position is compared with the target position determined in the preoperative planning, and the Euclidean distance deviation between the two is calculated. If the calculated distance deviation is less than the preset arrival judgment threshold, an arrival prompt is output, indicating to the doctor that the current catheter tip has reached the surgical lesion position and the subsequent surgical operation can begin. If the distance deviation is still greater than the arrival judgment threshold, the aforementioned fast loop-slow loop alternation process continues to complete pose estimation, drift compensation, and calibration parameter updates, continuously outputting the latest accurate pose result until the catheter tip is in place.

[0062] The generated navigation guidance prompts can be output to the display terminal to help doctors grasp the relative position of the current catheter position to the target position in real time, guide doctors to complete the catheter advancement operation, and further improve the operation efficiency and accuracy of navigation surgery.

[0063] In addition to calculating Euclidean distance, the system can also calculate the coordinate deviation between the catheter tip position and the target position in various anatomical dimensions. These deviations are displayed on the navigation interface, allowing doctors to more intuitively understand the direction and distance that the catheter tip needs to be adjusted, further enhancing the intuitiveness of the navigation guidance. It can also calculate the path deviation distance between the actual position and the planned path, determining whether the overall shape of the current catheter deviates from the preset surgical path. If the deviation exceeds the safe range, a path deviation warning can be simultaneously output, prompting the doctor to adjust the advancement direction in time to avoid damage to normal tissue.

[0064] The arrival notification message can be presented in various forms such as text highlighting, color change, and notification sound. This application does not limit the specific output form of the notification message, as long as it can convey the arrival signal to the doctor.

[0065] based on Figure 1In another optional implementation of the illustrated embodiment, when calculating the pose estimation result of the endoscopic catheter based on multi-source sensor observation data, a preset initial base weight can be assigned to each sensor source. This initial base weight is pre-set according to the inherent accuracy and reliability of the sensor source, with shape sensing receiving the highest initial base weight. For example, the shape sensing fiber itself can directly acquire the three-dimensional morphology of the catheter without relying on kinematic models and preoperative calibration parameters, and its pose observation accuracy is higher and its reliability is stronger; therefore, it receives the highest initial base weight. The encoder receives a second-highest initial base weight. If there are additional electromagnetic positioning or intraoperative imaging observation sources, corresponding weights are assigned according to their actual observation accuracy.

[0066] Furthermore, the consistency deviation between the observation results of each sensor source can be calculated, and the real-time weights of each sensor source can be adjusted based on the consistency deviation. The larger the consistency deviation, the greater the reduction in the real-time weight of the corresponding sensor source. This consistency deviation refers to the deviation between the pose observation result output by a single sensor source and the pose fusion result of all sensor sources. The larger the deviation, the lower the reliability of the current observation result of that sensor source, and therefore its current real-time weight should be reduced accordingly. The smaller the deviation, the larger its current real-time weight should be increased accordingly, thereby achieving dynamic adjustment of the weights of each sensor source.

[0067] For example, when the shape sensor outputs abnormal observation results due to local signal interference, the consistency deviation between its observation results and those of other sensor sources will significantly increase. The system will automatically reduce the real-time weight of the shape sensor and increase the weight ratio of other sensor sources such as the encoder to avoid the abnormal sensor observations having an excessive impact on the final fusion result. When the interference disappears, the consistency deviation decreases, and the weight of the shape sensor will automatically recover to a higher level, making full use of its high-precision observation advantage. This dynamically adapts to the real-time working state of different sensor sources, minimizing the interference of abnormal sensor observations on the pose estimation results and further improving the stability and accuracy of the pose estimation results. This mechanism of dynamically adjusting weights based on consistency deviation can adapt to the observation reliability of each sensor source under different operating conditions, further improving the robustness of multi-source fusion pose estimation.

[0068] Subsequently, the observation results from each sensor source can be further weighted and fused based on the adjusted real-time weights to obtain the pose estimation result of the endoscopic catheter. For example, incremental weighted Kalman filtering can be used to fuse multi-source observation results. First, the observation results from each sensor source are used as independent observation inputs. The corresponding observation noise covariance is assigned according to the real-time weights. Then, the optimal pose estimation result is obtained through filtering and iterative updates. This approach can combine the observation advantages of different sensor sources and suppress the influence of abnormal observations through dynamic weights.

[0069] based on Figure 1In another optional implementation of the illustrated embodiment, the pose estimation results are corrected and compensated. One approach is to calculate the cumulative drift correction amount for the current fast loop cycle based on the difference between the calibration parameters updated in the previous fast loop cycle and the initial calibration parameters obtained from the preoperative one-time calibration, combined with the cumulative drift trend. Specifically, a trend fitting model can be used to fit the differences in calibration parameters recorded in multiple historical slow loop cycles to obtain a predictive model of the calibration parameter deviation changing over time. This model is then used to calculate the total deviation of the calibration parameters corresponding to the current fast loop cycle, and based on this total deviation, the cumulative drift correction amount that needs to be superimposed on the pose estimation for the current fast loop cycle is calculated. This approach can fully utilize the trend information accumulated from historical drifts to achieve advance compensation for the drift in the current cycle. Even if the parameter update for the current slow loop cycle has not yet been triggered, the current pose result can be compensated in advance based on the existing drift trend, further improving the timeliness of drift compensation throughout the entire process.

[0070] Furthermore, the cumulative drift correction can be superimposed on the pose estimation result to obtain the corrected pose result. Specifically, based on the mapping relationship between calibration parameter deviation and pose error, the total deviation of calibration parameters can be directly converted into a cumulative drift correction in the pose space. This correction is then added to the original pose estimation result of the current fast cycle to obtain the corrected pose output result. If the original pose estimation result is an uncompensated pose obtained through multi-source fusion, the converted cumulative drift correction can be directly superimposed on the position and attitude components of the original pose estimation result to obtain the final pose result compensated for the effects of chronic drift, which can be used for navigation output and subsequent residual calculation.

[0071] This method enables continuous advance compensation for the persistently accumulated chronic drift during the parameter update interval between two adjacent slow loop cycles, preventing the drift error from accumulating continuously within the slow loop update interval and further improving the accuracy and stability of the pose estimation throughout the entire process.

[0072] Subsequently, the pose correction result can be weighted and filtered based on the current multi-source fusion confidence level to remove abnormal fluctuation components. The current multi-source fusion confidence level is calculated based on the uniformity of the real-time weight distribution of each sensor source. For example, if the real-time weight distribution of each sensor source is more concentrated on a single sensor source, and the overall level of consistency deviation is higher, the multi-source fusion confidence level is lower, corresponding to a stronger filtering and smoothing intensity, to filter out abnormal fluctuation components in the correction result. Conversely, if the weight distribution is more dispersed and balanced, and the overall level of consistency deviation is lower, the multi-source fusion confidence level is higher, corresponding to a weaker filtering and smoothing intensity, preserving the detailed dynamic changes in the pose result. This ensures both the stability of the pose result and the response speed of pose tracking.

[0073] The final pose correction result after weighted filtering and smoothing can be used as the compensated pose estimation result of the final output of the current fast loop cycle, and used in subsequent navigation guidance and drift discrimination processes.

[0074] Of course, the filtered pose correction result can be further subjected to amplitude limiting processing to restrict the maximum variation of a single correction, thus obtaining the compensated pose estimation result. In other words, it is possible to...

[0075] The difference between the cumulative drift correction of the current fast loop cycle and the cumulative drift correction of the previous fast loop cycle is calculated. If the difference exceeds the preset maximum single correction change threshold, the difference is clipped to the maximum allowable range. The cumulative drift correction of the current fast loop cycle is then calculated based on the clipped difference and re-superimposed to obtain the pose correction result. This avoids excessive single correction amplitude causing pose output jumps, further ensuring the stability of navigation output and avoiding interference with the doctor's operation.

[0076] based on Figure 1 In another optional implementation of the illustrated embodiment, one approach to determining the predicted pose of the endoscopic catheter is to use the compensated pose estimation result output from the previous fast loop cycle, combined with the catheter joint control input received in the current fast loop cycle, and based on a pre-stored endoscopic catheter kinematic model, to calculate the theoretical prediction of the overall catheter shape and tip position for the current fast loop cycle, thus obtaining the predicted pose. The catheter joint control input refers to the joint bending, rotation, or propulsion displacement output by the doctor's operating handle or robot drive mechanism. This input drives the changes in the position and shape of the endoscopic catheter and can be used to predict the movement trends of the catheter tip and various joint segments.

[0077] The predicted pose, calculated by combining the kinematic model with the driving input (i.e., the control input of the catheter joint), can serve as prior information for multi-source fusion pose estimation. This helps correct the observation biases of each sensor source, further improving the accuracy of the pose estimation results. Especially when a sensor source experiences a brief observation anomaly, the predicted pose can provide reasonable prior constraints, avoiding large fluctuations in the pose results. The pre-stored endoscopic catheter kinematic model can be a DH kinematic model based on the rigid body assumption of the catheter segments, or a flexible deformation model fitted based on actual mechanical properties. This application does not limit the specific type of kinematic model, as long as it can be used to calculate the corresponding pose prediction result based on the driving input.

[0078] In this process, if the current fast-loop cycle is the first fast-loop cycle in the entire navigation process, the pose result after the previous successful navigation reset is used as the initial reference for predicting the pose. In other words, if the pose calculation for any fast-loop cycle has not been completed after the navigation reset, the reference calibration parameters and initial pose stored after the previous slow-loop cycle update are used as the initial reference for predicting the pose. The accuracy of the initial reference directly determines the starting accuracy of the pose estimation in the first fast-loop cycle. Therefore, after each navigation reset and slow-loop parameter update, the currently stable pose result is synchronously stored as the initial reference for the next prediction calculation, preventing the initial reference error from being amplified by subsequent processes and ensuring the cumulative accuracy of the pose estimation throughout the entire process.

[0079] based on Figure 1 In another optional implementation of the embodiment shown, when short-window residual features and long-window residual features are obtained by statistically analyzing multiple pose residuals based on multiple fast loop cycles, one approach is to select the pose residuals of the most recent N fast loop cycles to calculate the mean, variance, and maximum amplitude of the short-window residuals as short-window residual features, and select the pose residuals of the most recent M fast loop cycles to calculate the trend slope, cumulative integral, and average amplitude of the long-window residuals as long-window residual features, where M > N and all are positive integers.

[0080] Furthermore, when judging instantaneous anomalies based on the characteristics of short-window residuals, if the variance of the short-window residuals is greater than a preset variance threshold, or the maximum amplitude is greater than a preset residual amplitude threshold, then an instantaneous anomaly is determined to exist, and the number of instantaneous anomalies in the slow window of the fast loop cycle is accumulated.

[0081] When performing chronic drift discrimination based on long-window residual features, if the trend slope of the long-window residual is greater than a preset slope threshold, or the cumulative integral is greater than a preset integral threshold, or the average amplitude of the long-window residual is greater than a preset average amplitude threshold, it is determined that chronic drift has accumulated beyond the allowable range, and the calibration parameter update of the slow loop cycle needs to be triggered; if none of the above features exceed the corresponding threshold, the current calibration parameters remain unchanged, the slow loop update process is not triggered, and the pose estimation and compensation of the next fast loop cycle continues.

[0082] For example, if the short-window residual feature meets the instantaneous anomaly detection criteria, it indicates that there is a sudden abnormal interference in the current multi-source observation results. In this case, even if the long-window residual feature meets the drift triggering criteria, the calibration parameter update for this slow-loop cycle will not be triggered, thus avoiding erroneous residuals from abnormal observations from misleading the parameter update process. Only when the short-window residual feature does not meet the instantaneous anomaly detection criteria, and the trend slope of the long-window residual feature is greater than the preset drift slope threshold, or the cumulative integral is greater than the preset drift integral threshold, is it determined that there is a chronic cumulative drift that needs correction, triggering the slow-loop cycle to start the calibration parameter update process. This ensures the accuracy of the parameter update triggering criteria from the residual feature level and avoids false triggering.

[0083] By using a dual-layer feature discrimination mechanism that combines long and short windows, it can effectively distinguish between instantaneous observation anomalies and long-term cumulative drift. It will not misjudge instantaneous abnormal fluctuations as chronic drift and trigger unnecessary parameter updates, nor will it miss the cumulative trend of chronic drift. This ensures the accuracy of drift discrimination, reduces invalid slow loop calculations, and prevents drift accumulation from being undetected in a timely manner.

[0084] based on Figure 1 In another optional implementation of the illustrated embodiment, when updating the calibration parameters for pose estimation, the correction direction and correction step size of the calibration parameters can be obtained based on long-window residual feature fitting. Incremental adjustments are then made to the original calibration parameters to correct the deviation component caused by the accumulation of chronic drift. Specifically, a mapping relationship is established between the residual features and the calibration parameter correction amount. The direction of residual accumulation is directly mapped to the deviation direction of kinematic parameters or joint zero-position parameters in the calibration parameters to determine the correction direction. The correction step size is then determined based on the magnitude of the cumulative integral of the long-window residuals; the larger the cumulative residual, the larger the corresponding correction step size. This incremental adjustment method only makes small corrections to the original calibration parameters, without completely overturning the original parameter results. This avoids excessively large parameter update amplitudes that could cause a jump in pose results, ensuring the smoothness of the parameter update process. After correction, the updated calibration parameters are stored as the benchmark parameters for the next round of drift compensation and pose estimation, completing the parameter update process for this slow-loop cycle.

[0085] During the correction process, a maximum allowable range can be set for the correction amount of the calibration parameters, i.e., a preset maximum correction magnitude threshold. If the calculated calibration parameter correction amount exceeds this threshold, the correction amount is trimmed to the maximum allowable range. This avoids excessive single correction magnitudes that could cause the calibration parameters to deviate from a reasonable range, further ensuring the stability and reliability of the parameter update process. Specifically, the modulus of the calibration parameter correction amount can be calculated first. If the modulus is greater than the preset maximum correction magnitude threshold, the correction amount is scaled proportionally until the modulus equals the maximum correction magnitude threshold. Then, the scaled correction amount is superimposed on the original calibration parameters. This ensures timely correction of existing parameter deviations while preventing excessive single correction magnitudes from causing abnormal jumps in the calibration parameters, ultimately achieving smooth and reliable incremental parameter updates.

[0086] Optionally, after the correction is completed, the cumulative residual and long and short window residual features are reset simultaneously to provide an initial clean statistical benchmark for the next round of drift accumulation discrimination, and to avoid historical residual features interfering with the next drift trigger judgment.

[0087] If, after correction, verification reveals that the pose residual increases instead of decreasing, the parameter update is terminated, and the calibration parameters before the update are retained. In other words, the pose residual for the current fast-loop cycle is recalculated after correction. If the corrected pose residual is greater than the original pose residual, it indicates an error in the correction direction. In this case, the system reverts to the original calibration parameters before correction, and the correction result is not saved. This prevents erroneous correction from increasing the pose estimation error and ensures the overall stability of the calibration parameter updates. If the corrected pose residual decreases as expected, the updated calibration parameters are retained as the benchmark parameters for pose estimation and drift compensation in subsequent fast-loop cycles.

[0088] based on Figure 1 In another optional implementation of the illustrated embodiment, the updated calibration parameters are progressively updated into the pose estimation calculation of the first fast cycle after the slow cycle using a gated smooth write mechanism. One specific approach is as follows:

[0089] When the updated calibration parameters from the slow-loop cycle output are received, it is determined whether there are any unrecovered transient anomalies. An unrecovered transient anomaly refers to an abnormal state in which the short-window residual characteristics in the current fast-loop cycle exceed the anomaly threshold and have not returned to the normal range. For example, if the maximum amplitude of the short-window residuals in the most recent K consecutive fast-loop cycles exceeds the preset residual amplitude threshold, and the maximum amplitude of the short-window residuals in the current fast-loop cycle has not yet decreased below the threshold, or if the variance of the short-window residuals in the most recent K consecutive fast-loop cycles exceeds the preset variance threshold, and has not yet returned to below the variance threshold, then the observation anomaly is considered to have not been eliminated. The updated calibration parameters are not written temporarily, and the update operation is performed after the anomaly is eliminated to avoid introducing parameter update errors under abnormal conditions.

[0090] Therefore, if there is an unrecovered transient anomaly, the parameter update is suspended and the original calibration parameters are maintained until the anomaly is eliminated before the calibration parameter update is started. This is because the residual characteristics under transient anomaly interference are biased, and the calibration parameter correction results calculated based on this bias are less reliable. Directly writing them can easily introduce erroneous parameter corrections. Suspending the update can avoid the negative impact of erroneous corrections on subsequent pose estimation.

[0091] If no unrecovered transient anomalies exist, the system starts from the original calibration parameters and adjusts them towards the updated calibration parameters with a preset step size each time. This step size adjustment is completed once per fast loop cycle, and the parameter switching is gradually completed over multiple consecutive fast loop cycles. Specifically, the parameter difference between the original calibration parameters and the updated target calibration parameters is calculated and broken down into multiple sub-increments of preset step sizes. Only one sub-increment is added to the currently used calibration parameters in each fast loop cycle. Through gradual adjustments over multiple fast loop cycles, the currently used calibration parameters are smoothly transitioned to the updated target calibration parameters. This gradual switching method avoids pose output jumps caused by abrupt changes in calibration parameters, making the calibration parameter change process smoother and preventing abrupt interference with the doctor's operation, further improving the continuous stability of the navigation output.

[0092] For example, the number of sub-increments that need to be split between the original calibration parameters and the target parameters is positively correlated with the magnitude of the difference between the original calibration parameters and the target parameters. The larger the magnitude of the difference, the more sub-increments are split, and the more fast loop cycles are needed to complete the smooth transition. The smaller the magnitude of the difference, the fewer sub-increments are split, and the parameter switching can be completed faster, taking into account both the timeliness of parameter updates and the stability of the output process.

[0093] Optionally, if a new transient anomaly is detected during the gradual switching process, the parameter switching process is paused, the current intermediate parameters are maintained unchanged, and subsequent gradual adjustments are resumed after the anomaly is eliminated, ensuring that the parameter switching process is not disturbed by transient anomalies. In other words, if a transient anomaly persists and cannot return to the normal range for an extended period, the current parameter update is terminated, the updated target calibration parameters are discarded, and the currently adjusted intermediate parameters are retained, or the original calibration parameters are rolled back directly, avoiding the introduction of additional errors by forcibly completing the parameter switch under unreliable observation conditions. If the transient anomaly is eliminated and recovery is possible, the remaining sub-incremental adjustments continue until a smooth transition to the target calibration parameters is achieved, thus keeping the entire parameter update process under control and further ensuring the continuous stability of the pose output.

[0094] Secondly, after each step size adjustment, the pose residual for the current fast loop cycle can be calculated. If the pose residual continues to decrease, the parameter update continues according to the preset step size. That is, if the pose residual shows a decreasing trend for several consecutive fast loop cycles, it indicates that the parameter adjustment direction is correct, and the adjustment continues until the entire switching process is completed. If the pose residual shows an abnormal increase, the current parameter adjustment is paused, and the parameter state of the previous adjustment cycle is returned to observe the current residual change. If the residual returns to normal after the return, it indicates that the step size adjustment range is too large, and the subsequent adjustment step size can be reduced before continuing. If the residual is still abnormal after the return, the current parameter update is terminated, and the currently working parameters are retained to avoid erroneous parameter adjustments from amplifying the pose estimation error.

[0095] During this parameter update process, if the pose residual calculated in the current fast loop cycle increases by more than a preset percentage threshold compared to the average pose residual of the most recent fast loop cycles before the update, or if the absolute value of the current pose residual exceeds the preset residual safety threshold, or if a new transient anomaly is detected, then this parameter update will be terminated, the original calibration parameters before the update will be rolled back, and the conservative control mode will be triggered.

[0096] In other words, when the above-mentioned abnormal situation occurs, it indicates that the target correction result of this parameter update is unreliable. Continuing to advance the parameter switching will only further amplify the pose estimation error. Therefore, this update is abandoned directly, and the original calibration parameters that have been working stably before the update are returned to. At the same time, a conservative control mode is activated: In the conservative control mode, the parameter update trigger judgment of all subsequent slow loop cycles is suspended, only the drift compensation calculation of the fast loop cycle is retained, and the correction step size is automatically reduced to a preset proportion of the normal step size. The pose residual statistical results of multiple fast loop cycles are continuously retained. After the pose residuals of multiple consecutive fast loop cycles have fallen below the safety threshold and no instantaneous abnormality is detected, the conservative control mode is exited and the normal slow loop parameter update trigger judgment process is restored. This minimizes the negative impact of erroneous parameter updates on the overall navigation process and ensures that the navigation process is always in a stable and controllable state.

[0097] For example, if the current pose residual increases by more than 30%, or the absolute value of the current residual exceeds the maximum error range allowed by normal navigation, it indicates that there is a deviation in the overall direction of this parameter update. Continuing to adjust will seriously affect the navigation accuracy. Therefore, it directly reverts to the original calibration parameters that have been verified to be reliable over a long period of time before the update, and triggers the conservative control mode. That is, in the subsequent K fast loop cycles, the instantaneous anomaly discrimination threshold is increased and the drift trigger sensitivity is reduced to avoid triggering unstable parameter updates again in a short period of time. The normal discrimination mechanism is restored after the observation state returns to stability, so as to maximize the safety and reliability of the navigation process.

[0098] based on Figure 1In another optional implementation of the illustrated embodiment, the computer device can further classify the system's operating status into several levels based on the current multi-source fusion confidence level, chronic drift level, and cumulative number of instantaneous anomalies: normal and reliable, low confidence warning, parameter update limit, parameter update freeze, navigation pause and rollback, and manual takeover. Furthermore, based on the current detection results matching the corresponding operating level, each operating level corresponds to a preset safety control strategy. For example, when entering the low confidence warning level, a drift warning is issued while maintaining the current navigation process, only increasing the confidence level of the pose result. Specifically, a prominent low confidence warning indicator is added to the pose display area of ​​the navigation interface to remind the operator that the confidence level of the current navigation result has decreased, requiring a judgment of operational safety based on clinical experience, without directly interrupting the navigation operation.

[0099] When entering the parameter update limit level, the single parameter correction amount and pose adjustment range are limited, while the residual detection frequency is increased. For example, the maximum correction range of the calibration parameters can be further reduced to half of the normal allowable range, limiting the magnitude of single parameter correction and reducing the fluctuation risk brought about by parameter updates.

[0100] When the navigation pause / retreat level is reached, the endoscopic catheter is retracted to the nearest confidence verification point to reconfirm its pose before further advancement. Specifically, if the cumulative pose residual at the nearest confidence verification point on the advancement path is below the safety threshold and there are no transient abnormal disturbances, pose estimation is restarted at that verification point. After confirming that the current pose matches the navigation output, the operation can continue. If the pose verification at that point fails, the catheter is retracted to the previous confidence verification point for re-verification until a verification point with the required confidence level is found before restarting navigation advancement. This avoids operational risks caused by continuing advancement when the pose deviation is too large. For example, the navigation process can be automatically paused, an alarm can be issued, and the operator can be guided to retract the instrument to a known reference position to complete the initial calibration, eliminate cumulative drift, and then continue the navigation operation.

[0101] When the parameter update freeze level is entered, all online parameter update operations are stopped, and the current calibration parameters are retained to complete the remaining navigation. For example, all slow loop cycle parameter update triggers can be stopped, and only the pose estimation calculation of the fast loop cycle can be retained. The parameter update function can be resumed after the observation state returns to stability.

[0102] When the manual takeover level is reached, the automatic control output is automatically cut off, control is handed over to the operator, and a full-process status log is recorded. In other words, when the above safety control strategies fail to bring the pose residual back to a safe range, the manual takeover level is entered, prompting the termination of automatic navigation and the operator to switch to manual operation mode. This establishes a multi-layered safety protection mechanism to ensure the operational safety of the entire navigation surgery process.

[0103] All level switching and safety operations are recorded with timestamps and corresponding trigger indicators, supporting subsequent traceability and verification. The system can record all relevant indicators for each level switching, including the trigger time, pose residual at the time of triggering, number of instantaneous anomalies, current drift level, and correction amount, storing them in the system's local log file. This facilitates post-operative review and analysis of the entire navigation process's state changes, and also allows for optimization of level classification thresholds and corresponding safety strategies for different scenarios, further enhancing the overall safety and adaptability of the system.

[0104] The following is through Figure 2 The timing diagram shown below further illustrates the endoscopic catheter calibration and drift compensation method of this application embodiment. As shown in the figure, the timing of this endoscopic catheter calibration and drift compensation method includes the following multiple stages:

[0105] During the startup and self-test phase, after the system is powered on, it first checks sensor communication, actuator status, and controller initialization in sequence, and then completes the initial reference calibration and parameter initial value setting. The system only enters the navigation operation state when communication, hardware status, and initial calibration values ​​are all normal. For example, the joint zero-position parameters and kinematic calibration parameters obtained from the initial calibration can be read first, and these parameters can be stored in the parameter cache as the currently effective calibration parameters. At the same time, the long and short window residual statistics cache, cumulative residual integral, and all status flags are initialized to ensure that all statistics are accumulated from the initial clean state, avoiding the impact of initial random interference on the accuracy of subsequent drift discrimination. After the self-test is completed and all status initializations are completed, the first fast-loop cycle of sensor acquisition and pose estimation is started, entering the continuous navigation operation phase.

[0106] During the dual-loop phase of operation, the fast loop is responsible for high-frequency execution control, while the slow loop is responsible for low-frequency parameter updates. Preferably, within a single fast loop cycle, data acquisition, fusion estimation, control command generation, and safety checks are completed first, and then the execution results are output. After every few fast loop cycles, the system enters a slow loop check to complete online calibration, drift trend calculation, and parameter update determination.

[0107] For example, a slow loop is triggered every N fast loop cycles, where N is a preset fixed number of cycles, or dynamically adjusted based on the current cumulative residual size. The larger the cumulative residual, the shorter the interval for triggering the slow loop, completing parameter correction earlier and preventing the continuous accumulation of drift. After entering the slow loop, the long and short window residual characteristics cached for the current N fast loop cycles are read first to determine whether chronic drift exceeding the threshold has accumulated: if the residual does not exceed the drift trigger threshold, the current slow loop is exited directly, and the fast loop cycle continues to run without parameter updates; if the residual exceeds the drift trigger threshold, the online calibration calculation of the slow loop is started to obtain the new calibration parameter correction amount and the updated target calibration parameters. Then, according to the aforementioned anomaly check and gating smooth write mechanism, the parameters are gradually updated into the pose estimation process. After completing this round of drift compensation, the slow loop is exited, and the normal fast loop operation is resumed, waiting for the next slow loop trigger.

[0108] Throughout the dual-loop operation, each fast loop cycle independently performs instantaneous anomaly detection and safety level determination, executing corresponding control operations according to preset safety strategies. This ensures timely response and navigation safety even if anomalies occur during slow loop parameter updates. When navigation operations are completed or termination conditions are triggered, the system exits the continuous navigation operation phase and enters the finalization phase. It stores the entire process log, clears parameter and statistical caches, shuts down sensor communication and actuator output, and completes the entire navigation calibration process.

[0109] During the gated smooth write phase, after the slow loop calculates the new calibration parameters, the system does not directly write them into the fast loop. Instead, it gradually transitions from the old parameters to the new parameters within a predetermined time window. During the transition period, the fast loop only senses continuous changes, thus avoiding actuator output jitter or overshoot caused by sudden parameter changes.

[0110] For example, the system might first detect any unrecovered transient anomalies. Only after confirming that the observation status is stable and there are no unresolved transient interferences will the gradual parameter transition begin. If unrecovered transient anomalies exist, the current parameter update will be suspended, maintaining the original calibration parameters unchanged, and the parameter update process will be restarted after the observation status stabilizes. After confirming that there are no anomalies, the difference between the original calibration parameters and the updated target calibration parameters is calculated. The number of sub-increments is determined based on the magnitude of the difference; the larger the magnitude, the more sub-increments are needed. Subsequently, only one sub-increment is added to the currently effective calibration parameters in each fast-loop cycle. The parameter switching is gradually completed within multiple consecutive fast-loop cycles to avoid abrupt parameter changes affecting the stability of the navigation output.

[0111] During the gradual switching process, instantaneous anomalies and pose residual changes are detected synchronously in each fast loop cycle. If a new instantaneous anomaly is detected, the switching process is paused, the current intermediate parameters are maintained unchanged, and the process continues only after the anomaly is eliminated. If the anomaly cannot be eliminated for a long time, the parameter update is terminated directly, and the original calibration parameters before the update are rolled back to avoid errors introduced by forced switching. After each step size adjustment, the pose residual of the current fast loop cycle is calculated. If the pose residual continues to decrease, it indicates that the adjustment direction is correct, and the process continues according to the preset step size. If the pose residual increases abnormally, the adjustment is paused, and the parameter state of the previous cycle is rolled back. If the residual returns to normal after the rollback, the subsequent adjustment step size is reduced and the process continues. If the residual is still abnormal after the rollback, the update is terminated directly, and the original normal parameters are retained.

[0112] If, during parameter switching, the pose residual increases beyond a preset proportional threshold or its absolute value exceeds a safety threshold, the entire update process is terminated immediately, reverting to the original calibration parameters before the update. Simultaneously, a conservative control mode is triggered, pausing subsequent slow-loop parameter update triggers and reducing the fast-loop drift compensation step size. Once the residual continuously falls back to a safe range and there are no abnormalities, the normal parameter update judgment process is resumed, ensuring that the entire parameter update process is stable and controllable, and avoiding interference from parameter mutations to navigation operations.

[0113] During the anomaly recovery and shutdown phase, if continuous drift deterioration, complete sensor loss, or simultaneous anomalies from multiple sources occur, the system immediately enters a degradation process. After the task is completed, the system stops pose updates and compensation outputs, saves calibration parameter history, drift trajectory, alarm status, and operation logs, and then releases resources to enter standby mode.

[0114] Specifically, in the case of continuous drift deterioration, if the pose residual still cannot fall back to the safe threshold after multiple parameter updates, it means that the current drift level has exceeded the range that a single parameter correction can solve. The system directly triggers a navigation pause prompt, guiding the operator to withdraw the instrument to the most recently verified reference pose point, complete the initial calibration at the reference point, obtain new calibration parameters, and then restart navigation to avoid the cumulative drift from causing navigation failure.

[0115] For scenarios where sensors are completely out of lock, if a sensor source fails to output valid data for multiple consecutive loop cycles, the fusion weights corresponding to that sensor source are frozen, and the system switches to a single-source or multi-source fusion estimation mode that relies solely on the remaining valid sensor sources. Simultaneously, a sensor anomaly alert is issued to remind operators to monitor the sensor status. If all sensor sources are out of lock, navigation termination is triggered, and manual takeover is initiated. For scenarios where multiple sources are simultaneously abnormal, if two or more sensor sources detect a transient anomaly or their output confidence level falls below a threshold, parameter update freezing is triggered. Only the loop pose estimation based on the original calibration parameters is retained, while the anomaly detection frequency is increased. The parameter update process resumes only after most sensor sources return to normal output and the confidence level returns to within acceptable limits.

[0116] After completing the aforementioned degradation process or the navigation task is completed normally, the system will store the update history of all calibration parameters during this navigation process, the pose residuals calculated in each fast loop cycle, the cumulative drift trajectory throughout the entire process, the time and corresponding level of all alarm triggers, and the execution records of all safety policies to the designated log file. Then, it will clear the parameter cache, residual statistics cache, and status identifier cache of the current cycle, shut down the continuous acquisition process of each sensor source, cut off the automatic control output, wait for the next navigation task start command, and enter the system standby state.

[0117] In the multi-source compensation method that uses shape sensing as the primary source and encoders as the secondary source, shape sensing is used as the primary observation source, encoders as the secondary verification source, and visual or image information as the low-frequency verification source. Specifically, shape sensing obtains the three-dimensional deformation profile of the entire duct by acquiring the wavelength signal of a fiber Bragg grating, directly outputting the absolute pose of each joint point. The encoder is installed at the drive joint of the duct, acquiring the motor rotation angle, and calculating the relative pose increment at the duct end through a kinematic model. This provides preliminary correction to the overall deformation error of the shape sensing. At preset intervals of shape sensing acquisition, the encoder outputs a cumulative angle increment, which is compared with the pose change obtained from the shape sensing for residual verification to determine whether slow drift accumulation has occurred in the shape sensing.

[0118] When endoscopic or intraoperative images are acquired during the operation, the positions of the marker points in the images are extracted and matched with the pose output by the current navigation system to complete the low-frequency global verification, further correct the long-term drift accumulation error, and finally integrate the three types of sensor information through the multi-source fusion framework to obtain a more accurate and stable full-process pose estimation result.

[0119] The system can also fuse pose using Kalman filtering or equivalent methods, and complete online calibration and drift compensation through a dual-loop structure. This scheme has a simple structure and is suitable for scenarios with low visual dependence. For example, the Kalman filtering framework uses the absolute pose output from the shape sensor as the measurement update term and the relative pose increment output from the encoder as the prediction update term. In each fast loop cycle, the state prediction is first completed based on the encoder increment, and then the state covariance is updated by combining the measurement results from the shape sensor to obtain the current optimal pose estimation result. After entering the slow loop process at each preset fast loop cycle, the filter residual statistics of the most recent multiple cycles are extracted to determine whether there is chronic drift in the calibration parameters. If the drift exceeds the threshold, the online optimization and adjustment of the calibration parameters are initiated, and the adjusted parameters are updated into the kinematic model and the shape sensor compensation model to achieve continuous drift compensation and correction.

[0120] During the filtering process, the measurement noise covariance is dynamically adjusted based on the real-time confidence of different sensor sources. When the output confidence of a certain sensor source decreases, the weight of that sensor source in the filtering update is automatically reduced, and the contribution of sensor sources with higher confidence to the results is retained first, thus ensuring the stability of the fusion results.

[0121] In some alternative approaches, factor graphs or sliding window optimization can be used to replace probabilistic filtering. This involves replacing the Kalman filter in the multi-source fusion module with factor graph optimization, sliding window optimization, or other methods suitable for strongly nonlinear scenarios to improve estimation accuracy under complex bending and strongly nonlinear stress conditions. In factor graphs, the multi-source pose measurements for each fast-loop cycle are used as an observation factor, and the relative motion increment between adjacent cycles is used as an odometry factor. This constructs an optimization problem containing multiple historical window pose states. By continuously removing old historical states and adding new observations through a sliding window, the computational cost of the optimization window remains constant. Each time a slow loop is triggered, the state and calibration parameters within the window are jointly optimized to obtain the current optimal calibration parameter correction. This approach can better handle estimation errors caused by strongly nonlinear deformation, further improving drift compensation accuracy. This method is more suitable for platforms with high computing power.

[0122] Furthermore, adaptive thresholding and adaptive gating can be employed to dynamically adjust drift thresholds, confidence thresholds, and gating thresholds based on noise levels, changes in operating conditions, or surgical stages, allowing the system to adopt different safety strategies at different stages. This approach enhances system adaptability. For example, during the initial calibration stage, immediately after instrument introduction and before deep manipulation, drift is minimal, allowing for a more appropriate increase in the drift trigger threshold to reduce unnecessary parameter update frequency. As deep manipulation progresses, increased instrument deformation and accelerated drift accumulation necessitate an automatic decrease in the drift trigger threshold, triggering more frequent slow-loop parameter updates to promptly correct drift and prevent error accumulation. Finally, during the surgical closure stage, only minor adjustments to the end-effector pose are required, automatically raising the safety trigger threshold to minimize unnecessary pauses and ensure smooth operation.

[0123] When adjusting noise levels, if environmental interference is strong and the overall variance of sensor output noise increases, the amplitude threshold for instantaneous anomaly detection is automatically increased to prevent normal noise from being misjudged as anomaly and triggering unnecessary safety interventions. When environmental interference decreases and sensor output stabilizes, the anomaly detection threshold is automatically reduced to improve anomaly detection sensitivity.

[0124] In other alternative approaches, a distributed estimation and control method can be employed, decentralizing some estimation capabilities to the duct tip or local modules. The host computer handles global coordination and safety management, thereby reducing communication latency and improving local response speed. Specifically, a micro-acquisition and processing unit can be integrated locally at the duct tip to directly acquire, preprocess, and perform preliminary deformation calculations of shape sensing signals. Only the calculated joint pose results are transmitted to the host computer, eliminating the need to transmit the original high-dimensional wavelength acquisition data and significantly reducing the data bandwidth usage of bus communication. The local drive module integrates encoder signal acquisition and incremental calculation units locally, directly outputting the processed angle increment results, reducing the sampling and calculation burden on the host computer, while shortening the latency of local anomaly response and ensuring faster execution of local safety strategies.

[0125] The host only needs to complete multi-source pose fusion, global drift determination, parameter update and optimization, and overall security management. The overall computing power allocation is more balanced, the communication latency is lower, and it can support higher frequency fast loop control, further improving the real-time response performance of the navigation system.

[0126] Beyond shape sensing and encoders, compensation methods based on image or end-vision verification can be used to enhance the role of image matching or end-vision verification in parameter update triggering. This allows the slow loop to determine the existence of true drift based on richer environmental observations. For example, using intraoperative images obtained by intraoperative endoscopy or CBCT as a global reference, the actual coordinates of the instrument's built-in markers in the image are extracted and registered with the coordinates of the markers estimated by the current navigation system. The offset of the navigation result relative to the actual position is calculated, and this offset is converted into a correction amount for calibration parameters. This is directly used as the constraint input for the slow loop's online calibration calculation, correcting potential misjudgments that may arise from relying solely on residual statistics to determine drift, and improving the accuracy of online calibration.

[0127] If an end-effector vision module is configured during surgery, the intraoperative field of view captured by the end-effector camera is directly extracted. Features of fixed anatomical landmarks within this field of view are extracted and matched with corresponding features in the preoperative planning model to obtain the true error of the current navigation pose. This error is then input into the slow-loop calibration and optimization process to obtain a calibration correction result that better reflects the actual intraoperative situation, further improving the accuracy of navigation throughout the procedure. This approach is more suitable for surgical scenarios with open intraoperative imaging support, fully utilizing existing intraoperative imaging equipment to complete high-frequency calibration and further suppressing long-term drift accumulation.

[0128] Furthermore, risk prediction triggering can be used instead of fixed threshold triggering. This replaces the fixed threshold drift triggering mechanism with triggering conditions based on trend prediction, risk scoring, or stage context, allowing for earlier detection of deteriorating trends and reducing false triggers. In other words, the drift trend is fitted based on the residual statistical results of multiple historical periods to predict when the drift will reach the risk threshold. If the predicted drift will exceed the safe range within a preset operating period, the slow-loop parameter update is triggered in advance, avoiding excessive correction due to the drift accumulating to a dangerous level before adjustment.

[0129] Simultaneously, the system generates a drift risk score based on the current surgical stage, instrument stress state, and sensor noise level. Slow-loop updates are only triggered when the score reaches a preset level, reducing invalid triggers in drift-free scenarios and minimizing interference with normal operation from parameter updates. When triggering based on stage context, slow-loop correction is only triggered during stages where drift is more likely to occur and the instrument posture is relatively stable, such as instrument retraction or reversal operations. During continuous, uniform instrument advancement, the original calibration parameters are maintained, balancing the need for correction accuracy with operational smoothness.

[0130] Regardless of the fusion algorithm, threshold strategy, communication link, or hardware deployment method used, the technical essence of the embodiments of this application remains unchanged: during the operation of the endoscopic robot catheter, pose and drift are continuously estimated using multi-source data, and the combination of short-window anomaly and long-window trend is used to determine whether the online calibration parameters need to be updated. Through the decoupling of fast loop and slow loop and the gating smooth writing mechanism, continuous drift compensation and safety degradation control are achieved without interrupting the operation.

[0131] The following uses an airway navigation scenario as an example to further illustrate in detail the embodiments of this application in which the computer device, based on the specific functions of its multiple functional modules, implements the above-described endoscopic catheter calibration and drift compensation method, and its various optional implementation methods. Please refer to... Figure 3 The computer equipment includes at least a sensor acquisition and time synchronization module, an online calibration parameter estimation module, a drift detection and decomposition module, a multi-source fusion state estimation module, a compensation controller, and a security monitoring and degradation management module. Its core architecture adopts a dual-loop decoupled design of "fast loop execution + slow loop update." Specifically:

[0132] Module 1: Sensor Acquisition and Time Synchronization Module. This module acquires multi-source data, including shape sensing, encoder data, image matching, and respiratory status data, and performs timestamp alignment, missing data detection, and basic preprocessing to form a synchronized data vector. Therefore, this embodiment does not simply rely on a single sensor, but rather allows multiple observation sources to work together under confidence constraints. Even if one source experiences interference or loss of lock at a certain stage, the system can still maintain pose estimation and compensation continuity by relying on other high-confidence sources. This advantage can be verified through pose variance and control stability under electromagnetic interference, visual occlusion, or local deformation scenarios.

[0133] Module 2: Online Calibration Parameter Estimation Module. This module continuously estimates the sensing and control mapping parameters based on the fast-loop accumulated residuals and the current operating conditions, and outputs online updated calibration parameters. This module operates in the slow loop and does not directly rewrite the fast-loop control parameters at a high frequency; instead, it first performs a convergence check.

[0134] This application's embodiment employs a fast-loop control process, a slow-loop parameter update process, and a gated smooth write connection between the two. This prevents parameter updates from directly impacting the fast-loop control, significantly reducing the risks of overshoot, oscillation, and execution jitter. This advantage can be verified through evaluations of control variance, overshoot amplitude, and operator subjective stability.

[0135] Module 3: Drift Detection and Decomposition Module. This module constructs drift indices from the residuals between the fused pose and the predicted pose, and distinguishes between rapid abnormal disturbances and long-term drift trends. This module calculates the residual fluctuation amplitude within a short time window and the residual accumulation trend within a long time window, thereby differentiating between instantaneous rapid disturbances caused by breathing motion and long-term slow drifts caused by calibration parameter offsets. This avoids misjudging instantaneous disturbances as drifts requiring compensation, improving the stability of compensation.

[0136] Preferably, this module uses both short and long time windows: the short window is used to capture transient anomalies, such as sensor lock-up, rapid contact events, or respiratory mutations. Rapid contact events refer to a sudden change in posture caused by the catheter tip or sidewall momentarily contacting the airway wall, airway tissue, or surrounding adjacent tissue during catheter advancement. The posture change caused by this type of event is a transient, non-cumulative, one-time disturbance, unlike the slow, cumulative drift caused by friction or elastic relaxation. This module only briefly marks such transient disturbances and does not classify them as long-term drifts requiring calibration parameter updates, thus avoiding erroneous triggering of slow-loop parameter updates and resulting control deviations. The module also adjusts the fusion confidence weights of multi-source data at the corresponding time point based on the marked transient disturbance information to prevent disturbance data from interfering with the state estimation results. The long window is used to identify whether the residual sequence is continuously increasing, thereby determining whether there is a true drift requiring online calibration updates.

[0137] The embodiments of this application can distinguish between short-term anomalies and chronic drift, and will not misjudge single respiratory fluctuations, short-term exposures, or occasional noises as true drift requiring parameter updates, thereby reducing false triggers and overcorrection. This advantage can be verified by the false alarm rate, true drift detection rate, and parameter update effectiveness. The combined discrimination of short-window anomalies and long-window trends is more suitable for high-noise medical environments.

[0138] Module 4: Multi-source fusion state estimation module, used to output conduit pose estimation and covariance based on multi-source observations and the confidence levels of each source. Unlike traditional methods that only weight based on noise magnitude, this application embodiment preferably considers the following three confidence dimensions simultaneously: signal quality, such as noise variance, signal-to-noise ratio, and observation availability; inter-source consistency, i.e., whether the observation deviation between different sources is within the allowable range; and temporal synchronization, i.e., whether the timestamp delay between different sources is within the allowable range.

[0139] For example, firstly, the signal quality confidence of each sensor source is calculated based on the raw acquired data of the sensor signals. The larger the noise variance and the lower the signal-to-noise ratio, the lower the signal quality confidence. If a sensor source has no effective output, the signal quality confidence is set to zero. Secondly, based on the pose observation results of each sensor source's current output, the deviation between the observation results of each pair of sensor sources is calculated. If the deviation exceeds the allowable range, the inter-source consistency confidence of the sensor source with the larger deviation is reduced. Finally, based on the deviation between the timestamp of each sensor source's data and the system's master clock, it is determined whether the timing synchronization error exceeds the tolerance. If it does, the timing synchronization confidence of that sensor source is reduced. Finally, the confidence of the three dimensions is weighted and fused to obtain the overall confidence of the sensor source, which serves as the basis for adjusting the weights during the fusion process, thereby improving the robustness of the multi-source fusion results in dealing with disturbances in complex scenarios.

[0140] Module 5: Compensation Controller. This module generates feedforward and feedback micro-compensation commands based on the compensated pose estimate, the current task objective, and historical control states. The controller preferably operates in a fast loop to maintain real-time control consistency. The task objective could be, for example, guiding the catheter tip towards the pre-planned target lesion location, or performing advancement, retraction, or angle fine-tuning operations while maintaining pose stability, according to operator-given adjustment commands. The compensation commands output by this module automatically limit the adjustment range to prevent excessive single-compensation amplitude from causing catheter jitter or tissue displacement, ensuring a smooth and controllable operation.

[0141] Module Six: The Safety Monitoring and Degradation Management module is used to perform graded actions such as gating, limiting, freezing, rollback, or manual takeover on the outputs of Modules Two and Five based on drift indicators, drift trends, fusion confidence levels, and anomaly durations. This module also undertakes the control responsibility for the smooth writing of slow-loop parameters to the fast-loop.

[0142] Therefore, through the aforementioned functional modules, online calibration can be achieved, significantly improving navigation accuracy throughout the entire process. Continuous updating of key mapping parameters via online recalibration allows parameter biases to be corrected promptly during the procedure, eliminating the need for passive corrections at critical points. The direct result is suppressed cumulative error growth, leading to smaller and more stable endpoint deviations in long-path navigation. This advantage can be verified through the relationship between endpoint error and path length, parameter convergence speed, and consistency of repeated trials.

[0143] Furthermore, this application's embodiments directly incorporate drift indicators and confidence levels into security gating, enabling tiered execution of low-confidence, limit, freeze, rollback, and manual takeover based on the severity of anomalies, thereby prioritizing security boundaries when the system is unreliable. This advantage can be verified through anomaly detection rate, response time, recovery success rate, and log traceability. Tiered security degradation enhances system controllability and compliance verifiability.

[0144] In some alternative approaches, some estimation and compensation capabilities can be offloaded to the duct tip or end-effector module, while global coordination is performed by the host. Specifically, the duct tip integrates a micro-processing unit that directly completes the preprocessing steps such as raw acquisition of shape sensing signals, noise filtering, and coordinate calculation. It only packages and sends the calculated pose results of each joint segment to the host, eliminating the need to transmit raw, high-dimensional fiber optic wavelength acquisition data. This significantly reduces the data bandwidth usage of bus communication between the master and slave devices, while also reducing the basic computational load on the host, freeing up more computing power for global fusion and security management.

[0145] For the instrument's built-in encoder module, signal acquisition and incremental calculation units can be integrated locally on the control board of the local drive joint. This allows for the direct output of processed joint angle increments, transmitting only the final incremental data to the host computer instead of the original coded pulse signals. This further reduces the host computer's sampling and calculation burden, while also shortening the response latency for local anomalies, enabling faster execution of local safety strategies and ensuring the safety of local operations. The host computer only needs to perform multi-source pose fusion, global drift trend determination, calibration parameter optimization and updates, and overall surgical safety management. The system's computing power distribution is more balanced, communication latency is lower, and it can support higher-frequency fast-loop control cycles, further improving the real-time response performance of the navigation and control system and adapting to higher-precision operational requirements.

[0146] High-resolution external images can also be used as the primary observation source, with sensors serving only as auxiliary calibration tools. For example, CBCT, fluoroscopy, or endoscopic images captured in real-time during surgery can be used as a global pose reference. The actual pixel coordinates of instrument-borne markers or feature points in the images can be extracted, projected into three-dimensional space, and registered and aligned with the corresponding point coordinates estimated by the current navigation system. The offset of the navigation pose relative to the actual position can be calculated, and this offset can then be converted into a correction constraint for calibration parameters. This constraint is then input into the slow-loop online calibration optimization process to correct the misjudgment problem that easily occurs when relying solely on residual statistics to determine drift, thereby further improving the accuracy of online calibration results.

[0147] If an end-effector vision module is configured during surgery, the intraoperative visual field features captured by the end-effector camera are directly extracted and matched with fixed anatomical landmarks in the preoperative planning images to obtain the true error of the current navigation pose. This error is then input into the slow-loop calibration optimization process to obtain calibration correction results that better reflect the actual intraoperative scenario, further suppressing long-term drift accumulation and improving the positioning accuracy of navigation throughout the entire procedure. This image verification method can fully utilize existing intraoperative imaging equipment to complete high-frequency or periodic calibrations, making it more suitable for open surgical scenarios that support real-time intraoperative imaging, and can improve navigation accuracy without adding additional hardware configurations.

[0148] In alternative approaches to multi-source fusion, extended Kalman filtering, particle filtering, factor graphs, sliding window optimization, learned estimators, or other equivalent methods can be employed. Specifically, based on a pre-defined system model and observation equations, the state mean and covariance are continuously updated recursively. This approach is computationally efficient and easy to deploy in real-time, especially in scenarios with low state dimensions and noise that approximates a Gaussian distribution. When using particle filtering, a large number of weighted particles approximate the posterior probability distribution, enabling the application of non-Gaussian and nonlinear observation scenarios and providing better robustness to conditions with complex deformations and high model uncertainty.

[0149] When using factor graph optimization, the observations from each sensor source are treated as different factors to construct a unified graph optimization problem. The optimal pose estimation result is obtained through incremental iteration, which can effectively utilize historical observation information and achieve better long-term estimation consistency. When using sliding window optimization, only historical observation data within a fixed-length window is retained, which controls the amount of computation while maintaining optimization accuracy and is more suitable for intraoperative navigation scenarios with long-term continuous operation. When using a learning estimator, the end-to-end estimation model is trained based on a large amount of labeled data collected offline. When running online, the fused pose result can be output by directly inputting multi-source raw data. It can fit complex nonlinear errors and simplify the model design process.

[0150] The above methods do not change the core inventive concept of dual-ring decoupling online calibration and multi-source joint compensation for drift in this application, and all fall within the scope of protection claimed in this application.

[0151] Alternative methods for drift detection include simple thresholding, integral triggering, adaptive thresholding, or risk prediction mechanisms. For example, simple thresholding involves setting a fixed residual accumulation threshold. When the accumulated residual over a long window exceeds this threshold, a slow-loop parameter update is directly triggered. This method is logically simple, requires low computational power, and is suitable for simple surgical scenarios with high real-time requirements and minimal scene disturbances. Integral triggering involves integrating and accumulating the residual excess over multiple periods. An update is only triggered when the integral value reaches a preset trigger threshold, avoiding direct updates from a single, accidental residual excess and further reducing the probability of false triggers. This method is suitable for surgical scenarios with high noise levels and a tendency for sporadic residual fluctuations.

[0152] As mentioned earlier, the adaptive threshold automatically adjusts the drift trigger threshold based on the current surgical stage, sensor noise level, and instrument operation status, balancing trigger frequency and operational smoothness. This provides greater adaptability and is suitable for complex, multi-stage interventional surgical scenarios. Risk prediction triggering, on the other hand, fits drift change trends based on historical residual data, predicting in advance when drift will exceed the safety threshold. It triggers slow-loop parameter updates before the drift reaches a dangerous level, avoiding large deviations caused by correction delays. This is suitable for drift control requirements in long-duration, complex surgeries. All of the above triggering methods fall within the scope of this application and can be freely selected and adapted according to the actual surgical scenario requirements.

[0153] The key judgment conditions involved in the embodiments of this application include: whether the end-to-end delay of the fast loop processing meets the real-time requirements; whether the long window residual trend is clearly increasing; whether the parameter change amount has reached a level worthy of updating; whether the drift index continuously exceeds the limit; and whether the timestamp delay of each source is within the synchronous tolerance range. Among them, the end-to-end delay of the fast loop processing refers to the time consumed by the entire link from the completion of sensor acquisition to the output of the compensation command. If this time exceeds the preset real-time tolerance threshold, ticket caching and degradation processing are triggered, and the current slow loop parameter update is not started. The real-time performance of the fast loop control is prioritized to avoid parameter updates competing for computing power and causing control lag.

[0154] Whether the residual trend within the long window is clearly upward is determined by linearly fitting the slope of the residual data across multiple consecutive frames within the long window. If the fitting slope is greater than a preset slope threshold and the goodness of fit meets the requirements, a clear long-term drift trend is identified, meeting the prerequisite for triggering parameter updates. Whether the parameter change is significant enough to warrant an update is determined by comparing the newly estimated calibration parameters with the old parameters currently being used in the fast loop, calculating the percentage change in magnitude. If the change is less than a preset minimum update threshold, the parameter write is not triggered, avoiding control fluctuations caused by minor and meaningless parameter updates.

[0155] Whether the drift metric continuously exceeds the limit can be determined by counting the number of consecutive frames where the drift metric exceeds the safety threshold. An update is only triggered when the number of consecutive frames exceeding the threshold reaches a preset count threshold, filtering out occasional abnormal fluctuations in the drift metric and reducing the probability of false triggers. Whether the timestamp delay of each source is within the allowable synchronization range is determined by comparing the deviation between the timestamp of each sensor source and the system master clock. If the delay of any sensor source exceeds the allowable range, the fusion confidence weight of that sensor source is reduced. If the delay exceeds the maximum allowable range, the confidence of that source is directly set to zero to eliminate interference from out-of-sync data on the fusion result.

[0156] The computer device in the embodiments of this application is described below. Please refer to [link / reference]. Figure 4 One embodiment of the computer device in this application includes:

[0157] The computer device 400 may include one or more central processing units (CPUs) 401 and a memory 405, in which one or more applications or data are stored.

[0158] The memory 405 can be volatile or persistent storage. The program stored in the memory 405 can include one or more modules, each module including a series of instruction operations on the computer device. Furthermore, the central processing unit 401 can be configured to communicate with the memory 405 and execute the series of instruction operations stored in the memory 405 on the computer device 400.

[0159] Computer device 400 may also include one or more power supplies 402, one or more wired or wireless network interfaces 403, one or more input / output interfaces 404, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0160] The central processing unit 401 can perform the aforementioned... Figure 1 The operations performed by the computer device in the illustrated embodiments and their various alternative embodiments are not described in detail here.

[0161] This application also provides a computer storage medium, one embodiment of which includes: the computer storage medium storing instructions, which, when executed on a computer, cause the computer to perform the aforementioned... Figure 1 The operations performed by the computer device in the illustrated embodiments and their various alternative embodiments.

[0162] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0163] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0164] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0165] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0166] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for calibrating and compensating for drift of an endoscopic catheter, characterized in that, The method includes: In each fast loop cycle, acquire multi-source sensor observation data for the endoscopic catheter; The pose estimation result of the endoscopic catheter is calculated based on the multi-source sensor observation data, and the pose estimation result is corrected and compensated to obtain the compensated pose estimation result. Determine the predicted pose of the endoscopic catheter, and determine the pose residual of the fast loop cycle based on the predicted pose and the compensated pose estimation result; After a preset number of fast loop cycles, a slow loop cycle is entered. In the slow loop cycle, short window residual features and long window residual features are statistically obtained based on the pose residuals of multiple fast loop cycles. The short window residual features are used for instantaneous anomaly detection, and the long window residual features are used for chronic drift detection. If the long window residual feature is determined to exceed the drift threshold based on the chronic drift discrimination, and the number of short window residual abnormalities is determined not to exceed the preset abnormality number threshold based on the instantaneous anomaly discrimination, then the calibration parameters of the pose estimation are updated to obtain the updated calibration parameters. The updated calibration parameters are gradually updated into the pose estimation calculation of the first fast cycle after the slow cycle through a gated smooth write mechanism; The short-window residual features and long-window residual features are statistically obtained from the multiple pose residuals based on multiple fast loop cycles, including: The mean, variance, and maximum amplitude of the pose residuals for the short window calculation over the most recent N fast loop cycles are selected as the features of the short window residuals. The trend slope, cumulative integral, and average amplitude of the pose residuals for the long window calculation over the most recent M fast loop cycles are selected as the features of the long window residuals, where M > N and all are positive integers. The step of using the short-window residual features for instantaneous anomaly detection and the long-window residual features for chronic drift detection includes: If the variance of the short window residual is greater than the preset variance threshold, or the maximum amplitude is greater than the preset residual amplitude threshold, then a transient anomaly is determined to exist, and the number of transient anomalies in the slow window of the fast loop cycle is accumulated. If the trend slope of the long window residual is greater than the preset slope threshold, or the cumulative integral is greater than the preset integral threshold, or the average amplitude of the long window residual is greater than the preset average amplitude threshold, then chronic drift is determined to exist. The updated calibration parameters are progressively updated into the pose estimation calculation of the first fast cycle after the slow cycle using a gated smooth write mechanism, resulting in the updated pose estimation results, including: When the updated calibration parameters are received from the slow loop cycle output, it is determined whether there is an unrecovered transient anomaly; the unrecovered transient anomaly refers to an abnormal state in which the short window residual feature exceeds the anomaly threshold and has not returned to the normal range within the current fast loop cycle. If the aforementioned unrecovered transient anomaly exists, the current parameter update will be suspended, and the original calibration parameters will be maintained until the anomaly is eliminated before the calibration parameter update is initiated. If there is no unrecovered transient anomaly, then starting from the original calibration parameters, the original calibration parameters are adjusted to the updated calibration parameters by a preset step size each time. The step size adjustment is completed once in each fast loop cycle, and the parameter switching is gradually completed within multiple consecutive fast loop cycles.

2. The method according to claim 1, characterized in that, The method further includes: Obtain target location information, which is used to indicate the target location that the tip of the endoscopic catheter needs to reach, as determined by the preoperative planning path. After each fast loop cycle pose estimation calculation is completed, based on the compensated pose estimation result and the target position information, it is determined whether the distance deviation between the current position of the tip of the endoscope catheter and the target position is within a preset range. If so, then determine that the endoscopic catheter has reached the target position, and output the compensated pose estimation result and the arrival prompt information; If not, continue with the pose estimation and pose compensation process for the next fast loop cycle.

3. The method according to claim 1, characterized in that, The calculation of the pose estimation result of the endoscopic catheter based on the multi-source sensor observation data includes: Each sensing source is assigned a preset initial base weight, which is preset based on the inherent accuracy and reliability of the sensing source, with shape sensing assigned the highest initial base weight. Calculate the consistency deviation between the observation results of each sensor source, and adjust the real-time weight of each sensor source according to the consistency deviation, wherein the larger the consistency deviation, the greater the reduction in the real-time weight of the corresponding sensor source. The observation results of each sensor source are weighted and fused based on the adjusted real-time weights to obtain the pose estimation result of the endoscopic catheter.

4. The method according to claim 1, characterized in that, The step of correcting and compensating the pose estimation result to obtain the compensated pose estimation result includes: Based on the difference between the calibration parameters updated in the previous fast loop cycle and the initial calibration parameters obtained from the preoperative one-time calibration, the cumulative drift correction amount for the current fast loop cycle is calculated in combination with the cumulative drift trend. The cumulative drift correction is superimposed on the pose estimation result to obtain the pose correction result; The pose correction result is weighted and filtered based on the current multi-source fusion confidence level to remove abnormal fluctuation components; the current multi-source fusion confidence level is calculated based on the uniformity of the distribution of the real-time weights of each sensor source. The filtered pose correction result is subjected to amplitude limiting processing to limit the maximum variation of a single correction, thereby obtaining the compensated pose estimation result.

5. The method according to claim 1, characterized in that, The updated calibration parameters for the updated pose estimation are obtained by updating the calibration parameters, including: Based on the long window residual feature fitting, the correction direction and correction step size of the calibration parameters are obtained, and incremental adjustments are made on the basis of the original calibration parameters to correct the deviation components caused by the accumulation of chronic drift in the original calibration parameters. During the correction process, a maximum allowable range is set for the correction amount of the calibration parameters. If the verification after correction finds that the pose residual increases instead of decreasing, the parameter update is terminated and the calibration parameters before the update are retained.

6. The method according to claim 1, characterized in that, The method further includes: Based on the current multi-source fusion confidence level, chronic drift level, and cumulative number of instantaneous anomalies, the system's operating status is divided into the following levels: normal and reliable, low confidence warning, parameter update limit, parameter update freeze, navigation pause and rollback, and manual takeover. Based on the current detection results, the corresponding work level is matched, and each work level corresponds to a preset security control strategy: When the low confidence warning level is reached, a drift warning is issued and the current navigation process is maintained, only increasing the confidence level of the pose result; When entering the parameter update limiting level, the single parameter correction amount and pose adjustment range are limited, while the residual detection frequency is increased. When the parameter update freeze level is reached, all online parameter update operations are stopped, and the current calibration parameters are retained to complete the remaining navigation. When the navigation pause and retraction level is entered, control the endoscope catheter to retract to the nearest confidence verification point, reconfirm the pose, and then advance it; When the system enters the manual takeover stage, it automatically cuts off the automatic control output, hands over control to the operator, and records the entire process status log.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 6.

8. A computer storage medium, characterized in that, The computer storage medium stores instructions that, when executed on the computer, cause the computer to perform the method as described in any one of claims 1 to 6.

Images