Nerve block puncture path dynamic planning system based on AI

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an AI-based nerve block puncture path dynamic planning system. The system comprises a data sensing synchronization unit used for collecting medical image data and respiratory signals of a patient in real time and synchronizing the time of the medical image data and the respiratory signals; the dynamic modeling unit is used for constructing and continuously updating an environment model containing a dangerous area and a dynamic change characteristic of a target nerve along with a respiratory cycle; the intelligent planning unit is used for generating and dynamically optimizing a puncture path from a puncture starting point to a target nerve through a path planning algorithm based on the environment model and the current breathing phase; the guide execution unit is used for providing multi-mode interaction feedback to assist puncture operation execution. The system can construct and update an environment model containing a dangerous area and a target nerve which change along with respiratory movement in real time, so that path planning is based on a real dynamic anatomical scene, and the safety and real-time adaptability of path planning are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an AI-based neural block puncture path dynamic planning system. BACKGROUND

[0002] The success of nerve block anesthesia highly depends on the accurate injection of anesthetics around the target nerve. Currently, ultrasound images are mainly relied on for real-time guidance in clinical practice. However, this operation has significant challenges: first, it is extremely dependent on the experience and hand-eye coordination of the operator, as the doctor needs to interpret two-dimensional ultrasound images, judge deep anatomical structures, and control the puncture needle to avoid blood vessels, pleura and other dangerous areas, with a steep learning curve and high operating pressure. Second, respiratory motion causes periodic displacement of the target nerve and surrounding tissue structures (such as the lung and diaphragm), making the puncture path planned based on static images invalid in actual operation, increasing the risk of injury and the uncertainty of operation.

[0003] To address these challenges, existing technologies have introduced some computer-aided planning methods, such as three-dimensional path static planning based on single-frame CT or MRI images, but these methods have the following shortcomings: 1. Unable to handle dynamic changes caused by respiratory motion, the planned path may become unsafe during inspiration or expiration.

[0004] 2. Although some studies have attempted to use respiratory gating technology to only advance the needle during a specific respiratory phase, this limits the timing of the operation and does not address the continuous movement of the target during the needle's travel within the tissue.

[0005] Therefore, an AI-based neural block puncture path dynamic planning system is designed. SUMMARY

[0006] The present application aims to provide an AI-based neural block puncture path dynamic planning system to solve the problems raised in the background.

[0007] To achieve the above-mentioned purpose, the present application aims to provide an AI-based neural block puncture path dynamic planning system, comprising: a data perception synchronization unit, configured to collect and time-synchronize medical image data and respiratory signals of a patient in real time; a dynamic modeling unit, configured to construct and continuously update an environment model containing the dynamic change characteristics of dangerous areas and target nerves with the respiratory cycle based on the synchronized medical image data and respiratory signals of the patient; An intelligent planning unit for generating and dynamically optimizing a puncture path from a puncture starting point to a target nerve by a path planning algorithm based on the environment model and its current respiratory phase; A guidance execution unit for fusing and presenting the planned path, real-time respiratory phase information, and the patient's safety state with the ultrasound image, and providing multi-modal interactive feedback to assist the puncture operation execution.

[0008] As a further improvement of the technical solution, the dynamic modeling unit includes a lightweight deep learning segmentation network, a time series prediction model, an online parameter learning module, and a safety mapping module; The lightweight deep learning segmentation network is used to perform real-time segmentation on the medical image data to identify the target nerve and dangerous tissue structures. The online parameter learning module is used to solve the specificity parameters of the current patient in the time series prediction model by fitting algorithm during the puncture preparation stage; The time series prediction model is used to predict the displacement change of the target nerve through a parameterized function; the parameterized function at least includes a periodic term related to the respiratory phase and patient-specific parameters; The safety mapping module is used to generate hierarchical safety regions with different safety levels based on the output of the segmentation network according to a plurality of preset safety distance thresholds, wherein the boundaries between each safety region are set as a probability transition zone.

[0009] As a further improvement of the technical solution, in the online parameter learning module, the specific steps for solving the specificity parameters of the current patient in the time series prediction model by fitting algorithm are as follows: S11, collect synchronous data of the patient's complete respiratory cycle; S12, use the lightweight deep learning segmentation network to extract the position sequence of the target nerve in the image coordinate system for the medical image data sequence; S13, perform band-pass filtering and peak detection on the respiratory signal to extract the instantaneous respiratory phase sequence; S14, solve the parameter set that minimizes the difference between the position sequence and the position prediction value calculated according to the time series prediction model by least squares fitting.

[0010] As a further improvement of the technical solution, the online parameter learning module is provided with a parameter real-time fine-tuning function, which continuously updates the parameters in a sliding window manner during the puncture process, and the window length is the data of the last respiratory cycles.

[0011] As a further improvement of the technical solution, the intelligent planning unit includes a global path planning module, a re-planning trigger module, and a local path re-planning module; The global path planning module is configured to plan an initial global path from the puncture starting point to the target nerve based on the initial environment model. The re-planning trigger module is configured to continuously compare the deviation of the actual puncture state from the planned path and monitor the uncertainty of the environment model, and trigger local re-planning when one of the preset conditions is met. The local path re-planning module is configured to respond to the re-planning trigger and perform incremental path planning based on the updated environment model within a limited time using an improved rapid-exploration random tree algorithm to generate a corrected path segment.

[0012] As a further improvement of the technical solution, the global path planning module uses a path planning algorithm based on graph search, which considers path length, risk integral, and synchronization cost related to respiratory phase in its cost function. Meanwhile, the global path planning module is configured to generate multiple candidate paths with different optimization objectives based on the same environment model and provide corresponding visualized and differentiated displays.

[0013] As a further improvement of the technical solution, the re-planning trigger module has the following specific conditions: Condition one: the deviation of the actual needle tip position from the planned path exceeds a dynamically set safety threshold; Condition two: the estimated uncertainty of the location of the target nerve or dangerous area in the environment model exceeds a set threshold; Condition three: the predicted displacement amplitude of the target nerve within the next respiratory cycle exceeds a tracking error threshold.

[0014] As a further improvement of the technical solution, the local path re-planning module generates a corrected path segment with the following specific steps: S21, problem definition: the module takes the three-dimensional position and attitude of the current needle tip as the planning starting point, and the planning endpoint is determined as follows: If the trigger reason is condition one and the latter path of the original global path is in a safe area, the planning endpoint is the nearest reachable point on the original global path; Otherwise, the planning endpoint is a local sub-target point dynamically calculated based on the current environment model; S22, constraint space construction: based on the real-time updated environment model of the dynamic modeling unit, a local three-dimensional search space centered on the current needle tip is delineated, and the boundaries of this space are constrained according to the distribution of dangerous areas and predicted displacement; The updated environment model includes the target nerve, dangerous areas, and their uncertainties; S23, rapid sampling planning: within the constraint space, an improved rapid-exploration random tree algorithm is used for path exploration, which includes a sampling bias strategy for needle instrument kinematics and a heuristic cost function for dynamic obstacles.

[0015] S24, path optimization and output: the initial path obtained by sampling is optimized for smoothness, and the corrected path segment is output, which is seamlessly connected with the current needle tip state and consistent in tangent direction.

[0016] As a further improvement of the technical solution, the guidance execution unit comprises a breath-synchronized guidance interface and a safety feedback module. The breath-synchronized guidance interface is used to provide visual indications of breathing phases and recommended operation timing in the display interface. The safety feedback module is used to generate feedback signals according to the real-time relationship between the current needle tip position and the planned path and the position of dangerous tissue structures.

[0017] As a further improvement of the technical solution, the feedback signals include visual feedback signals and auditory feedback signals.

[0018] Compared with the prior art, the beneficial effects of the present application are: In the AI-based neural block puncture path dynamic planning system, through the cooperation of the data perception synchronization unit and the dynamic modeling unit, an environment model containing dangerous areas and target nerves that change with respiratory motion can be constructed and updated in real time, so that the path planning is based on a real dynamic anatomical scene, fundamentally improving the safety and real-time adaptability of the planned path.

[0019] In the AI-based neural block puncture path dynamic planning system, the intelligent planning unit not only generates an initial global path, but also can quickly generate a corrected path when the needle tip deviates or the environmental prediction uncertainty increases through a triggered local re-planning mechanism, thereby significantly improving the robustness and fault tolerance of the puncture process and reducing the requirement for the operator's instantaneous error correction ability. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The overall flowchart of the present application is shown. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] Embodiment: Please refer to Figure 1 As shown, an AI-based neural block puncture path dynamic planning system is provided, comprising a data perception synchronization unit, a dynamic modeling unit, an intelligent planning unit and a guidance execution unit. Among them, the data sensing and synchronization unit is used to collect and time-synchronize the patient's medical imaging data and respiratory signals in real time; In practice, the operator uses an ultrasound device to continuously scan the target area (such as the supraclavicular region), and the system simultaneously acquires ultrasound image data streams and respiratory signals captured by a respiratory sensor.

[0023] The dynamic modeling unit is used to construct and continuously update an environmental model that includes the dynamic changes of danger zones and target nerves with the respiratory cycle, based on synchronized patient medical imaging data and respiratory signals. Danger zones refer to anatomical structures that need to be avoided during puncture, including but not limited to blood vessels (arteries and veins), pleura, peritoneum, non-target branches in nerve plexuses, and bones.

[0024] The target nerve refers specifically to the neural structure that needs to be blocked, such as the brachial plexus trunk, femoral nerve, sciatic nerve, etc. It is the endpoint of this system planning.

[0025] The dynamic modeling unit includes a lightweight deep learning segmentation network, a temporal prediction model, an online parameter learning module, and a secure mapping module; Among them, a lightweight deep learning segmentation network is used for real-time segmentation of medical image data to identify target neural and dangerous tissue structures; a U-Net variant based on MobileNetV3-small is adopted, with fewer than 100 parameters. With an inference time of less than 30ms per frame on edge computing devices, it can complete pixel-level segmentation of target nerves and dangerous tissues in a single frame of ultrasound image within 50ms.

[0026] The online parameter learning module is used to collect about 1 minute of the patient's natural breathing data during the puncture preparation stage, and solve the patient-specific parameters in the time-series prediction model through a fitting algorithm. In the online parameter learning module, the specific steps for solving the patient-specific parameters in the time-series prediction model using a fitting algorithm are as follows: S11. Collect synchronous data of the patient's complete respiratory cycle; S12. Use a lightweight deep learning segmentation network to extract the position sequence P(t) of the target nerve in the image coordinate system from the medical image data sequence. S13. Perform bandpass filtering and peak detection on the respiratory signal to extract the instantaneous respiratory phase sequence. ; S14. Solve for the parameter set that minimizes the difference between the location sequence and the location prediction value calculated based on the time series prediction model by least squares fitting. ; There are huge differences in patient's anatomy, breathing pattern, tissue elasticity. An "average" prediction model trained on population data may produce centimeter-level errors on a specific patient, which is unacceptable for nerve block puncture targeting millimeter-level accuracy. The online parameter learning module quickly "calibrates" the dynamic model parameters specific to the patient by using the patient's own real-time data during the preoperative preparation phase (preoperative planning / puncture).

[0027] The minimum difference parameter set enables the time series prediction model to achieve the most accurate prediction for the current patient, that is, to find the set of parameters that best explain the current observed patient's real motion data. This set of parameters makes the theoretical model match the patient's actual physiological response best.

[0028] Further, the online parameter learning module includes fault tolerance and adaptive mechanisms, which perform corresponding operations when the following abnormal conditions are detected during the process of solving the time series prediction model specific to the current patient: If the signal-to-noise ratio of the respiratory signal is lower than the threshold , prompt "poor signal quality, please adjust the sensor" and pause calibration; If the confidence of the segmentation network for consecutive N frames (N=10) of images is lower than the threshold , automatically switch to manual assisted segmentation mode; If the fitting residual exceeds the clinical tolerance error , discard the current data segment and reacquire; The online parameter learning module is provided with a parameter real-time fine-tuning function, which continuously updates the parameters in a sliding window manner during the puncture process, with an update frequency of 1 Hz and a window length of data of the last respiratory cycles.

[0029] The time series prediction model is used to predict the displacement change of the target nerve through a parameterized function; the parameterized function at least includes a periodic term related to the respiratory phase and patient-specific parameters; The displacement change of the target nerve is specifically:

[0030] In the formula, is the total displacement at time t; k is the anatomical coupling coefficient; A is the displacement amplitude; is the respiratory phase corresponding to time t; B is the baseline offset; C is the static offset; is a correction term to compensate for transient disturbances not covered by the model; The safety mapping module is used to generate hierarchical safety regions with different safety levels based on the output of the segmentation network according to a plurality of preset safety distance thresholds, wherein the boundaries between each safety region are set as a probability transition zone; Based on the segmentation result, the system generates three layers of safety zones: Absolute safety zone (green): >5mm from artery, >3mm from pleura; Cautious operation zone (yellow): 3-5mm from artery, 2-3mm from pleura; Forbidden zone (red): <3mm from artery, <2mm from pleura; Considering the resolution limit of ultrasound images, a probabilistic transition zone is added to each boundary; Safety probability ; where d is the actual distance, is the threshold distance, is the transition width parameter.

[0031] The traditional way is to set a hard threshold (e.g. 5mm), distances greater than 5mm are absolutely safe, and less than 5mm are absolutely dangerous. This can easily produce false positives (repeated alarms near the threshold) or false negatives in clinical practice.

[0032] The probabilistic transition zone is to establish a "gray area" (e.g. 4mm-6mm) near the hard threshold. In this area, the safety probability smoothly transitions from "dangerous" (probability 0) to "safe" (probability 1).

[0033] This can reflect the doctor's "more and more cautious" decision-making process when approaching the dangerous area, making the cost function of path planning continuous, the planned path more natural, and avoiding unreasonable sharp turns at the threshold. Reduces the risk of false triggering due to image noise or minor segmentation errors.

[0034] The intelligent planning unit is used to generate and dynamically optimize the puncture path from the puncture starting point to the target nerve based on the environment model and its current respiratory phase through the path planning algorithm; The intelligent planning unit includes a global path planning module, a re-planning trigger module, and a local path re-planning module; The global path planning module is used to plan an initial global path from the puncture starting point to the target nerve based on the initial environment model; The global path planning module uses a graph search-based path planning algorithm, which considers path length, risk integral, and synchronization cost related to respiratory phase in its cost function. At the same time, the global path planning module is configured to generate multiple candidate paths with different optimization objectives based on the same environment model, and provide corresponding visualized and differentiated display.

[0035] The system generates 1-3 candidate paths, displayed in different colors: Preferred path (blue solid line): optimal comprehensive score; Backup path (blue dashed line): safer but longer; Fast path (green dashed line): shortest but slightly riskier.

[0036] The re-planning trigger module is used to continuously compare the deviation of the actual puncture state from the planned path and monitor the uncertainty of the environment model, and trigger local re-planning when one of the preset conditions is reached; In the re-planning trigger module, the preset conditions are as follows: Condition one: the deviation of the actual position of the needle tip from the planned path exceeds the dynamically set safety threshold; the safety level is dynamically set to 2.0 mm (safe zone) or 1.0 mm (danger zone); the safety threshold is dynamically adjusted according to the distance between the current position of the needle tip and the dangerous tissue structure; Specifically, during the puncture process, the system calculates the Euclidean distance between the needle tip and the nearest dangerous tissue structure (such as the subclavian artery, pleural line) in real time.

[0037] According to the clinical safety specifications and the resolution of the ultrasound image (about 1 mm), the system sets two levels of safety threshold: When the Euclidean distance is greater than 5 mm, it is defined as "safe zone", and the allowed path tracking deviation threshold is set to 2.0 mm; When the Euclidean distance is less than or equal to 5 mm, it is defined as "alert zone", and the allowed path tracking deviation threshold is set to 1.0 mm.

[0038] The above threshold settings are based on: (1) 2.0 mm threshold: about 2 times the axial resolution of ultrasound, ensuring that small deviations within a safe distance do not trigger unnecessary re-planning, reducing operational interference; (2) 1.0 mm threshold: close to the limit of ultrasound resolution, providing the most stringent control when close to dangerous structures; (3) 5 mm demarcation line: based on clinical consensus, a distance of > 5 mm from the blood vessel is a relatively safe distance, which can avoid accidental contact caused by blood vessel pulsation or respiratory motion.

[0039] This setting can prevent path deviation caused by operator hand shaking or patient body position micro-motion; the dynamic threshold mechanism allows moderate flexibility in the safe area and provides strict protection in the dangerous area; avoids operational interruptions caused by frequent re-planning, and improves operational smoothness.

[0040] Condition two: the estimated uncertainty of the position of the target nerve or dangerous area of the environment model exceeds the set threshold; Specifically, the system estimates the position of the target nerve and dangerous tissue structure through an extended Kalman filter, and calculates the trace of its covariance matrix As a measure of uncertainty.

[0041] Setting uncertainty threshold (corresponding to standard deviation ).

[0042] When , the model estimation is considered unreliable, and needs to be recalibrated or data collected.

[0043] The threshold is set according to: (1) corresponding to the maximum displacement of the typical suprascapular brachial plexus nerve under quiet breathing; The confidence interval is about , which has approached the clinically acceptable positioning error limit; (2) When it exceeds this threshold, the safety of the path planning based on the estimation cannot be guaranteed.

[0044] Condition three: the predicted displacement amplitude of the target nerve in the next breathing cycle exceeds the tracking error threshold.

[0045] Setting the target nerve displacement tracking error threshold .

[0046] Calculate the maximum predicted displacement in the next breathing cycle , ∈ [current phase, current phase + 2π].

[0047] When , it is considered that the amplitude of respiratory motion is too large, and the system may not be able to complete tracking from the current position to the target position in the next cycle.

[0048] The threshold is set according to: (1) corresponding to the maximum displacement of the typical suprascapular brachial plexus nerve under quiet breathing; (2) When it exceeds this amplitude, the linearity assumption based on the current respiratory model may not hold, and needs to be re-evaluated.

[0049] This setting can prevent high-risk operations during periods of intense respiratory motion, prompting the operator to wait for a more suitable breathing phase (such as end-expiratory); ensure that the system works within the physiological dynamic range, avoiding over-capacity planning.

[0050] The local path re-planning module is used to respond to the re-planning trigger and, within the limited time, perform incremental path planning based on the updated environment model using the improved rapid exploration random tree algorithm to generate a corrected path segment. The limited time is .

[0051] The specific steps for the local path re-planning module to generate a corrected path segment are as follows: S21, problem definition: the module takes the three-dimensional position and attitude of the current needle tip as the planning starting point, and the planning endpoint definition is determined as follows: If the trigger cause is condition one and the post-global path of the original global path is in the safety region, plan the end point as the nearest reachable point on the original global path; Otherwise, plan the end point as a local sub-goal point dynamically calculated according to the current environment model; In S21, the determination method of the nearest reachable point is as follows: Along the original global path, search forward from the deviation point, for each path point , calculate the safety score of the straight line path from the current needle tip to , select the first point whose safety score exceeds the threshold S min and whose geometric distance is the shortest as the end point, wherein the safety score calculation considers the minimum safety distance on the path and the breathing phase synchronization.

[0052] The determination method of the local sub-goal point is as follows: Based on the latest environment model, a candidate point set is established in the safety region around the target nerve; Wherein, the generation of the candidate point considers: (1) The minimum distance to the dangerous tissue structure is greater than or equal to the safety threshold (3mm) (2) It is in the "stable window period" of the target nerve under the current breathing phase (3) The geometric constraint of being reachable from the current needle tip For each candidate point, calculate the cost function by considering distance, risk, breathing phase penalty, and direction adjustment amplitude, and select the candidate point with the minimum cost as the local sub-goal point.

[0053] S22, constraint space construction: based on the environment model updated in real time by the dynamic modeling unit, a local three-dimensional search space centered on the current needle tip is delineated, the boundary of which is constrained according to the distribution of dangerous regions and predicted displacement, so as to focus computing resources; The updated environment model includes the target nerve, dangerous regions and their uncertainties; S23, fast sampling planning: in the constraint space, an improved fast exploration random tree algorithm is used for path exploration, which specifically includes a sampling bias strategy for needle instrument kinematics and a heuristic cost function for dynamic obstacles.

[0054] S24, path optimization and output: smoothness optimization (using B-spline curve fitting) is performed on the initial path obtained by sampling to ensure its curvature continuity, which meets the physical motion characteristics of the puncture needle, and the corrected path segment is output, which seamlessly connects the current needle tip state and has consistent tangent direction, forming a smooth operation transition.

[0055] Wherein, the improved fast exploration random tree algorithm is configured as: Planning time upper limit milliseconds; Maximum number of sampling points ; The guidance execution unit is used to fuse the planned path, real-time respiratory phase information and the safety state of the patient with the ultrasound image, and provide multi-modal interactive feedback to assist the execution of the puncture operation.

[0056] The guidance execution unit includes a respiratory synchronization guidance interface and a safety feedback module; The respiratory synchronization guidance interface is used to provide visualized indications of the respiratory phase and recommended operation timing in the display interface; The safety feedback module is used to generate feedback signals according to the real-time relationship between the current needle tip position, the planned path and the position of the dangerous tissue structure.

[0057] The feedback signals include visual feedback signals and auditory feedback signals; Specifically, the visual feedback signals are real-time display of the current respiratory cycle phase and highlight display of the recommended puncture operation window period through the respiratory phase indication bar in the fused display interface; The auditory feedback signals provide non-verbal auditory streams that are coordinated with the visual feedback, including: (1) a periodic background prompt tone that changes in pitch synchronously with the respiratory phase, used to strengthen the operator's perception of the respiratory cycle; (2) a safety alarm sound that is triggered when the needle tip approaches or enters different levels of dangerous areas, and is distinguished from the background sound in rhythm or frequency.

[0058] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the present application.

Claims

1. An AI-based dynamic planning system for neural block puncture pathways, characterized in that, include: A data sensing and synchronization unit is used to collect and synchronize the patient's medical imaging data and respiratory signals in real time. A dynamic modeling unit is used to construct and continuously update an environmental model containing the dynamic changes of danger zones and target nerves with the respiratory cycle based on synchronized patient medical imaging data and respiratory signals. The intelligent planning unit is used to generate and dynamically optimize the puncture path from the puncture origin to the target nerve based on the environmental model and its current respiratory phase using a path planning algorithm. The guidance execution unit is used to fuse the planned path, real-time respiratory phase information and patient safety status with ultrasound images, and provide multimodal interactive feedback to assist in the execution of the puncture operation.

2. The AI-based dynamic planning system for nerve block puncture pathways according to claim 1, characterized in that: The dynamic modeling unit includes a lightweight deep learning segmentation network, a temporal prediction model, an online parameter learning module, and a secure mapping module; The lightweight deep learning segmentation network is used to segment the medical image data in real time to identify target nerves and dangerous tissue structures. The online parameter learning module is used to solve for the patient-specific parameters in the time-series prediction model during the puncture preparation phase using a fitting algorithm. A time-series prediction model is used to predict displacement changes of the target nerve using a parameterized function; the parameterized function includes at least a periodic term related to the respiratory phase and patient-specific parameters. The security mapping module is used to generate hierarchical security regions with different security levels based on the output of the segmentation network and according to multiple preset security distance thresholds, wherein the boundaries between each security region are set as probability transition zones.

3. The AI-based dynamic planning system for nerve block puncture pathways according to claim 2, characterized in that: In the online parameter learning module, the specific steps for solving the patient-specific parameters in the time-series prediction model using a fitting algorithm are as follows: S11. Collect synchronous data of the patient's complete respiratory cycle; S12. Use the lightweight deep learning segmentation network to extract the position sequence of the target nerve in the image coordinate system from the medical image data sequence; S13. Perform bandpass filtering and peak detection on the respiratory signal to extract the instantaneous respiratory phase sequence; S14. Solve for the parameter set that minimizes the difference between the position sequence and the position prediction value calculated according to the time series prediction model by least squares fitting.

4. The AI-based dynamic planning system for nerve block puncture pathways according to claim 3, characterized in that: The online parameter learning module includes a real-time parameter fine-tuning function, continuously updating parameters via a sliding window during the puncture process. The window length is the most recent... Data for one respiratory cycle.

5. The AI-based dynamic planning system for nerve block puncture pathways according to claim 4, characterized in that: The intelligent planning unit includes a global path planning module, a replanning triggering module, and a local path replanning module. Among them, the global path planning module is used to plan the initial global path from the puncture point to the target nerve based on the initial environment model; The replanning trigger module is used to continuously compare the deviation between the actual puncture state and the planned path, and monitor the uncertainty of the environmental model. When one of the preset conditions is met, local replanning is triggered. The local path replanning module is used to respond to replanning triggers and, within a limited time, performs incremental path planning using an improved fast exploratory random tree algorithm based on the updated environment model to generate corrected path segments.

6. The AI-based dynamic planning system for nerve block puncture pathways according to claim 5, characterized in that: The global path planning module adopts a graph search-based path planning algorithm. Its cost function comprehensively considers path length, risk integral, and synchronization cost related to breathing phase. At the same time, the global path planning module is configured to generate multiple candidate paths with different optimization objectives based on the same environment model and provide corresponding visual differentiation display.

7. The AI-based dynamic planning system for nerve block puncture pathways according to claim 6, characterized in that: The preset conditions in the replanning trigger module are as follows: Condition 1: The deviation between the actual position of the needle tip and the planned path exceeds the dynamically set safety threshold; Condition 2: The estimation uncertainty of the location of the target nerve or danger zone in the environmental model exceeds a set threshold; Condition 3: The predicted displacement of the target nerve in the next respiratory cycle exceeds the tracking error threshold.

8. The AI-based dynamic planning system for nerve block puncture pathways according to claim 7, characterized in that: The specific steps by which the local path replanning module generates the corrected path segment are as follows: S21. Problem Definition: The module uses the current three-dimensional position and orientation of the needle tip as the planning starting point, and the planning endpoint is defined as follows: If the triggering reason is condition one and the subsequent path of the original global path is in the safe zone, the planned destination is the nearest reachable point on the original global path; Otherwise, the planning endpoint is a local sub-target point dynamically calculated based on the current environment model; S22. Constraint Space Construction: Based on the environment model updated in real time by the dynamic modeling unit, a local three-dimensional search space centered on the current needle tip is defined. The boundary of this space is constrained according to the distribution of the danger zone and the predicted displacement. The updated environmental model includes target neural networks, danger zones, and their uncertainties; S23. Fast sampling planning: Within the constrained space, an improved fast exploration random tree algorithm is used for path exploration, specifically including a sampling bias strategy for the kinematics of the needle-like device and a heuristic cost function for dynamic obstacles. S24. Path Optimization and Output: Optimize the smoothness of the initial path obtained from sampling and output the corrected path segment, whose starting point is seamlessly connected to the current needle tip state and whose tangent direction is consistent.

9. The AI-based dynamic planning system for nerve block puncture pathways according to claim 8, characterized in that: The guidance execution unit includes a respiratory synchronization guidance interface and a safety feedback module; Among them, the respiratory synchronization guidance interface is used to provide a visual indication of the respiratory phase and recommended timing of operation in the display interface; The safety feedback module generates feedback signals based on the real-time relationship between the current needle tip position and the planned path and the location of hazardous tissue structures.

10. The AI-based dynamic planning system for nerve block puncture pathways according to claim 9, characterized in that: The feedback signals include visual feedback signals and auditory feedback signals.

Images