An in vitro magnetic target positioning and navigation method and system for the larynx

Abstract

The present invention discloses an extracorporeal magnetic target positioning and navigation method and system for the larynx, belonging to the field of medical equipment technology. The method comprises: constructing a three-dimensional anatomical model based on laryngeal imaging data and marking the position of a reference magnet. The optimal path of the catheter from the entrance to the target point is planned, with a built-in guide magnet at the front end and extracorporeal magnet traction control. The magnetic sensor array tracks the position in real time, and closed-loop control corrects deviations. The distance to the tissue is monitored in real time, and a warning is issued and the path is replanned when it is less than a threshold. The method of the present invention improves the accuracy and safety of catheter positioning and navigation.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical equipment, and in particular relates to an in vitro magnetic target positioning and navigation method for the larynx and a system thereof. Background Art

[0002] The larynx is a vital part of the human body for breathing and vocalization. Its complex internal structure and confined space encompass several key areas, including the epiglottis, vocal cords, and laryngeal ventricle. Accurately locating the target area is crucial for laryngeal procedures, such as laryngeal examinations, laryngeal tissue sampling, and laryngeal surgery. However, due to the unique structure of the larynx, traditional locating methods have numerous limitations.

[0003] Endoscopes are currently a commonly used tool for laryngeal operations, but they rely on direct vision. For some target areas located deep in the throat or in hidden areas, it may be difficult for the endoscope to directly observe them, resulting in inaccurate positioning. Secondly, endoscopic operation requires professional skills and high requirements for the operator. When operating in the narrow laryngeal passage, it may cause damage to surrounding normal tissues. Furthermore, the endoscopic field of view is easily interfered with by factors such as bleeding, secretions, or tissue edema, affecting the accuracy and reliability of positioning.

[0004] Regarding imaging localization, although imaging techniques such as CT and MRI can provide detailed information about the target area of ​​the larynx, they can only provide static images before surgery and cannot reflect dynamic changes during the operation in real time. Imaging examinations require the patient to cooperate and remain still, which is difficult for some patients who cannot tolerate long examinations (such as children or critically ill patients). Furthermore, imaging localization cannot guide the instrument to the target area in real time, requiring the operator to manually operate according to the imaging images, which increases the difficulty and error of positioning.

[0005] During laryngeal procedures, operators often need to manually manipulate instruments to reach the target area. This manual manipulation relies on the operator's experience and feel, making it difficult to guarantee accuracy and consistency from one procedure to the next. Manual manipulation is easily restricted by anatomical structures in the narrow laryngeal space, making it difficult for instruments to reach the target location and even potentially damaging surrounding normal tissue. Manual manipulation is also inefficient, especially for complex laryngeal target areas, requiring repeated adjustments to the instrument's position, increasing operation time and patient pain.

[0006] In summary, existing technologies for locating target areas in the larynx have many shortcomings. With the continuous development and improvement of magnetic target positioning and navigation technology, it is expected to provide a new solution for locating target areas in the larynx. Summary of the Invention

[0007] In order to solve the above technical problems, the present invention proposes an external magnetic target positioning and navigation method and system for the larynx to solve the problems existing in the above-mentioned prior art.

[0008] To achieve the above objectives, the present invention provides an in vitro magnetic target positioning and navigation method for the larynx, comprising the following steps:

[0009] Based on the laryngeal imaging data, a three-dimensional anatomical model of the larynx and target points is constructed;

[0010] placing a reference magnet at an anatomical landmark of the larynx, and marking position information of the reference magnet in the three-dimensional anatomical model;

[0011] Planning an optimal path for the catheter from the entrance to the target laryngeal point based on the spatial coordinates of the three-dimensional anatomical model;

[0012] A guide magnet is built into the front end of the catheter, and the external magnet is moved according to the optimal path. Based on the magnetic force and magnetic dipole torque generated by the external magnet acting on the guide magnet, the catheter is magnetically pulled and directional controlled;

[0013] A magnetic sensor array is used to track the position of the catheter tip in real time, and a closed-loop control algorithm is used to adjust the position of the external magnet to correct the catheter deviation.

[0014] Real-time acquisition of the distance between the catheter and surrounding tissues. When the real-time distance is less than the preset threshold, a contact warning is issued.

[0015] Pause the current navigation path and re-plan the catheter's path based on the collision warning until the catheter reaches the target location;

[0016] Based on a 3D anatomical model of the larynx and target, a deep learning model was used to obtain the optimal dose.

[0017] Based on the optimal dose, determining the laser light source power setting and the optical fiber output parameters;

[0018] Optimize the target position and irradiation sequence. When the catheter reaches the target position, irradiate the target by combining the laser light source power setting and fiber output parameters.

[0019] Optionally, the process of constructing a three-dimensional anatomical model of the larynx and the target point based on the laryngeal image data includes:

[0020] The laryngeal image data is segmented to extract the target target volume and laryngeal cavity structure, and a three-dimensional anatomical model of the larynx and the target target is constructed based on the target target volume and laryngeal cavity structure; the three-dimensional anatomical model includes the laryngeal anatomical contour and the spatial position and size shape of the target target.

[0021] Optionally, the magnetic sensor array includes several high-precision three-axis magnetic sensors, which are arranged around the throat to measure the magnetic fields generated by the reference magnet and the guide magnet. By analyzing the magnetic dipole field distribution, the three-dimensional coordinates and azimuth of the guide magnet relative to the reference magnet are calculated in real time by inversion, thereby obtaining the actual position information of the catheter.

[0022] Optionally, a magnetic sensor array is used to track the position of the front end of the catheter in real time, and deviation correction is performed through closed-loop control to ensure that the catheter reaches the target position. The process also includes:

[0023] The actual position information of the catheter is compared with the target position information corresponding to the optimal path to obtain a deviation vector. Based on the deviation vector, a PID closed-loop control algorithm is used to adjust the position and posture of the extracorporeal magnet to correct the direction of the catheter.

[0024] The present invention also provides an in vitro magnetic target positioning, navigation and directional irradiation system for the larynx, which is used to implement the method described, comprising: a model construction module, a path planning module, a magnetic control guidance module, a deviation correction module, a safety monitoring module, an intelligent rollback module, a dose calculation module, a light source and optical fiber calculation module and an irradiation planning module connected in sequence;

[0025] a model construction module, configured to construct a three-dimensional anatomical model of the larynx and target points based on the laryngeal image data, and mark the position information of a reference magnet in the three-dimensional anatomical model; wherein the reference magnet is placed at an anatomical landmark of the larynx;

[0026] A path planning module is used to plan an optimal path for the catheter from the entrance to the target point in the larynx in a three-dimensional anatomical model based on a path planning algorithm;

[0027] The magnetic control guidance module is used to embed a guide magnet at the front end of the catheter, move the external magnet according to the optimal path, and perform magnetic traction and directional control on the catheter based on the magnetic force and magnetic dipole torque generated by the external magnet acting on the guide magnet;

[0028] A deviation correction module is used to track the position of the front end of the catheter in real time using a magnetic sensor array and perform deviation correction through closed-loop control;

[0029] The safety monitoring module is used to obtain the distance information between the catheter and the surrounding tissue in real time. When the real-time distance is less than the preset threshold, a contact warning is issued;

[0030] The intelligent rollback module is used to pause the current navigation path and send a collision warning to the path planning module to replan the catheter's path to ensure that the catheter reaches the target location;

[0031] A dose calculation module, which uses a deep learning model to obtain the optimal dose based on a 3D anatomical model of the larynx and target.

[0032] a light source and optical fiber calculation module, connected to the dose calculation module, for determining the laser light source power setting and optical fiber output parameters based on the optimal dose;

[0033] The irradiation planning module is connected to the light source and fiber calculation module and the intelligent rollback module respectively, and is used to optimize the irradiation target position and irradiation sequence. When the catheter reaches the target target position, the target target is irradiated in combination with the laser light source power setting and the fiber output parameters.

[0034] Optionally, the model building module includes: a model building unit and a coordinate system establishing unit;

[0035] The model building unit is used to segment the laryngeal image data, extract the target target volume and laryngeal cavity structure, and build a three-dimensional anatomical model of the larynx and the target target based on the target target volume and laryngeal cavity structure;

[0036] The coordinate system construction unit is used to establish a three-dimensional coordinate system of the three-dimensional anatomical model, and to place a reference magnet at an anatomical landmark of the larynx and mark position information of the reference magnet in the three-dimensional coordinate system.

[0037] Optionally, the deviation correction module includes a real-time measurement unit and a deviation adjustment unit;

[0038] The real-time measurement unit is used to arrange a magnetic sensor array around the throat, measure the magnetic field generated by the reference magnet and the guide magnet, and calculate the three-dimensional coordinates and azimuth of the guide magnet relative to the reference magnet in real time by analyzing the magnetic dipole field distribution, thereby obtaining the actual position information of the catheter;

[0039] The deviation adjustment unit is used to compare the actual position information of the catheter with the target position information corresponding to the optimal path to obtain a deviation vector. Based on the deviation vector, a PID closed-loop control algorithm is used to adjust the position and posture of the extracorporeal magnet to correct the direction of the catheter.

[0040] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0041] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0042] The present invention also provides a computer program product, comprising a computer program, which implements the steps of the method when executed by a processor.

[0043] Compared with the prior art, the present invention has the following advantages and technical effects:

[0044] The present invention's in vitro laryngeal magnetic target positioning and navigation method achieves precise magnetic traction and directional control of the catheter by constructing a three-dimensional anatomical model of the larynx and target point. The method utilizes the magnetic forces of reference and guide magnets, combined with real-time tracking of a magnetic sensor array and a closed-loop control algorithm. Simultaneously, by monitoring the distance between the catheter and surrounding tissue in real time and issuing contact warnings, the navigation path can be adjusted promptly to ensure the catheter safely and accurately reaches the target point in the larynx. This improves the accuracy and safety of catheter positioning and navigation, successfully overcoming the difficulties presented by the complex anatomical structure of the larynx while avoiding damage to the normal laryngeal structure.

[0045] The illumination plan finally output by the present invention includes detailed parameters such as the required laser power, exposure time and irradiation position, and is combined with the navigation and positioning process to form a complete illumination planning program, which is not directly used as a treatment plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0047] Figure 1 is an overall flow chart of an embodiment of the present invention;

[0048] Figure 2 Schematic diagram of the system structure and preoperative modeling of an embodiment of the present invention;

[0049] Figure 3 This is a flowchart of the underlying path planning algorithm of an embodiment of the present invention;

[0050] Figure 4 This is a diagram showing the magnetic navigation control principle of an embodiment of the present invention. DETAILED DESCRIPTION

[0051] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0052] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0053] Example 1

[0054] This embodiment provides an in vitro magnetic target positioning and navigation method for the larynx, comprising the following steps:

[0055] Based on the laryngeal imaging data, a three-dimensional anatomical model of the larynx and target points is constructed;

[0056] placing a reference magnet A at an anatomical landmark of the larynx, and marking position information of the reference magnet A in the three-dimensional anatomical model;

[0057] Planning an optimal path for the catheter from the entrance to the target laryngeal point based on the spatial coordinates of the three-dimensional anatomical model;

[0058] A guide magnet B is built into the front end of the catheter, and an external magnet C is moved according to the optimal path. Based on the magnetic force and magnetic dipole torque generated by the external magnet C acting on the guide magnet B, the catheter is magnetically pulled and directional controlled.

[0059] A magnetic sensor array is used to track the position of the catheter tip in real time, and a closed-loop control algorithm is used to adjust the position of the external magnet C to correct the catheter deviation.

[0060] Real-time acquisition of the distance between the catheter and surrounding tissues. When the real-time distance is less than the preset threshold, a contact warning is issued.

[0061] Pause the current navigation path and re-plan the catheter's path based on the collision warning until the catheter reaches the target location;

[0062] Based on a 3D anatomical model of the larynx and target, a deep learning model was used to obtain the optimal dose.

[0063] Based on the optimal dose, determining the laser light source power setting and the optical fiber output parameters;

[0064] Optimize the target position and irradiation sequence. When the catheter reaches the target position, irradiate the target by combining the laser light source power setting and fiber output parameters.

[0065] A process of constructing a three-dimensional anatomical model of the larynx and a target point based on laryngeal image data, placing a reference magnet A at an anatomical landmark of the larynx, and marking position information of the reference magnet A in the three-dimensional anatomical model may include:

[0066] First, medical images are acquired and a three-dimensional model is constructed. Imaging techniques such as CT or MRI are used to obtain high-resolution image data of the patient's laryngeal anatomical structure and the location of the target. In this embodiment, the target is a laryngeal tumor. Thin-slice CT with a slice thickness of ≤1 mm is preferably used to clearly identify the various tissue layers of the larynx and the tumor boundary. The images are then segmented to extract the tumor volume and laryngeal cavity structure, and a digital anatomical model of the laryngeal region is constructed. This model contains information such as the laryngeal anatomical contours and the spatial location, size, and morphology of the tumor lesion. Anatomical landmarks (such as specific laryngeal cartilages or vertebrae) are selected within the model coordinate system to establish a surgical coordinate system, and the coordinates of the tumor lesion center and boundary are recorded.

[0067] At the same time, a reference magnet A is placed on the surface of the patient's neck skin near the anatomical landmark of the larynx. The position of the reference magnet A is synchronously acquired during the CT scan so that it appears in the image, thereby calibrating the position of the reference magnet A in the three-dimensional model. This step aligns the patient entity with the digital model: the reference magnet A serves as the origin of the coordinate system. Once the patient's position moves slightly, the navigation coordinates can be recalibrated by re-detecting the position of the reference magnet A. In this way, the constructed three-dimensional anatomical model not only provides a map of the cavity anatomy, but also establishes a one-to-one corresponding coordinate system between the model and the patient entity through the reference magnet A.

[0068] The process of planning an optimal path for a catheter from an entrance to a laryngeal tumor based on the spatial coordinates of the three-dimensional anatomical model may include:

[0069] Within a three-dimensional model of the laryngeal cavity, a feasible path is calculated from the oral laryngeal entrance to the tumor surface. The planning algorithm comprehensively considers the laryngeal cavity shape, tumor location, and catheter accessibility, avoiding anatomical obstacles and narrow bends. Algorithms such as A* or RRT can be used to search for an obstacle-avoiding path in three-dimensional space and discretize it into a series of waypoints. Path planning outputs a sequence of desired trajectory coordinates for catheter navigation. If the tumor is obscure or the passage is narrow, the algorithm plans a detour angle for the catheter to find a path that approaches the tumor.

[0070] It is feasible to embed a guide magnet B at the front end of the catheter, move the external magnet C according to the optimal path, and perform magnetic traction and directional control of the catheter based on the magnetic force and magnetic dipole torque generated by the external magnet C acting on the guide magnet B. The process includes:

[0071] The catheter is magnetically pulled and oriented using an external magnet C. External magnet C is a powerful magnet that is freely movable and directionally adjustable, such as a NdFeB permanent magnet with a diameter of 5 to 10 cm. The magnetic field generated by external magnet C acts on the guide magnet B built into the front end of the catheter, generating magnetic force and magnetic dipole torque, pulling the catheter and adjusting its orientation. The physical mechanism is as follows: The force acting on the guide magnet B in the magnetic field can be approximately expressed as the gradient force formula: , where m is the magnetic moment of guide magnet B, and guide magnet B is the magnetic field of external magnet C. This formula indicates that the field gradient of external magnet C exerts an attractive force on guide magnet B, causing it to move toward the direction of the stronger magnetic field. Furthermore, the magnetic moment acting on guide magnet B tends to align its magnetic moment with the direction of the external magnetic field, which can be used to adjust the catheter's posture: the magnitude of the moment satisfies . Therefore, by controlling the position of magnet C to generate an appropriate field gradient, guide magnet B and the catheter can be pulled toward the target. By rotating external magnet C to change the field direction, guide magnet B can be rotated to align with the target.

[0072] The magnetic control guidance process is carried out under the guidance of the navigation system. The external magnet C is moved according to the planned path, so that the guide magnet B gradually moves along the planned route. Specifically, the path data is received, the position and posture of the external magnet C are calculated according to the waypoint target, and then the control instruction of the external magnet C is output to the execution device. The execution device can be a robotic arm, which moves the external magnet C to the corresponding position and takes the corresponding orientation according to the instruction. At this time, the guide magnet B is moved in the desired direction by the magnetic force, and the doctor can push the catheter appropriately to make it move forward. Repeating this process, the front end of the catheter bypasses normal tissue and passes through narrow bends under the guidance of magnetic force, and finally reaches the tumor area. When the end of the catheter approaches and aligns with the tumor surface, the movement of the external magnet C is stopped and its position is fixed to stabilize the positioning of the catheter.

[0073] It is feasible to use a magnetic sensor array to track the position of the catheter tip in real time, and adjust the position of the external magnet C through a closed-loop control algorithm to correct the catheter deviation. The process of ensuring that the catheter reaches the tumor location includes:

[0074] This embodiment uses a magnetic sensor array to track the position of the catheter tip in real time and performs deviation correction through closed-loop control. The magnetic sensor array consists of multiple high-precision three-axis magnetic sensors arranged around the patient's throat to measure the magnetic field generated by the reference magnet A and the guide magnet B. By analyzing the magnetic dipole field distribution, the three-dimensional coordinates and azimuth of the guide magnet B relative to the reference magnet A coordinate system are calculated in real time. The typical positioning accuracy can reach 1-2mm, and the refresh rate is several 10Hz. Because it uses static magnetic field positioning, it is not blocked by the patient's body or interfered by metal surgical instruments, and has high robustness.

[0075] The actual position of the catheter tip is compared with the predetermined path or target position, and the positioning error (deviation vector) is calculated. Based on the deviation, a PID closed-loop control algorithm is used to adjust the position and posture of the external magnet C to correct the catheter's direction. The PID controller outputs the control increment based on the error calculation, and its control law is: ,in is the deviation of the catheter tip from the desired position, is the proportional, integral and differential gain. Control output Acting on the magnetic control guidance module, the position or angle of the external magnet C is corrected in real time to make the error approach zero.

[0076] For example, if the catheter is detected to be heading away from the tumor center, the rotation angle of external magnet C is increased to guide guide magnet B to realign the catheter. If the catheter is positioned too high or too low, the height of external magnet C is adjusted to pull the catheter back into position. Through continuous dynamic adjustments, the catheter smoothly advances along the planned path and accurately locates the target. Throughout the entire process, the distance between the catheter tip and surrounding tissue is monitored, indicating a potential contact risk if the catheter is too close, thereby ensuring safety.

[0077] This embodiment also provides an in vitro magnetic target positioning and navigation system for the larynx, for implementing the method described, comprising: a model building module, a path planning module, a magnetic control guidance module, a deviation correction module, a safety monitoring module, and an intelligent rollback module connected in sequence;

[0078] a model construction module, configured to construct a three-dimensional anatomical model of the larynx and the tumor based on the laryngeal image data, and mark the position information of a reference magnet A in the three-dimensional anatomical model; wherein the reference magnet A is placed at an anatomical landmark of the larynx;

[0079] A path planning module is used to plan an optimal path for the catheter from the entrance to the laryngeal tumor in a three-dimensional anatomical model based on a path planning algorithm;

[0080] The magnetic control guidance module is used to embed a guide magnet B at the front end of the catheter, move the external magnet C according to the optimal path, and magnetically pull and directional control the catheter based on the magnetic force and magnetic dipole torque generated by the external magnet C acting on the guide magnet B. Furthermore, the external magnet C is movable and positioned by a magnetic navigation device installed at the end of the robotic arm, and is controlled by instructions issued by the control / computing unit.

[0081] A deviation correction module is used to track the position of the front end of the catheter in real time using a magnetic sensor array and perform deviation correction through closed-loop control;

[0082] The safety monitoring module is used to obtain the distance information between the catheter and the surrounding tissue in real time. When the real-time distance is less than the preset threshold, a contact warning is issued;

[0083] The intelligent rollback module is used to pause the current navigation path and send a collision warning to the path planning module to replan the guide path to ensure that the catheter reaches the tumor location.

[0084] The model construction module can be implemented to include: a model construction unit and a coordinate system establishment unit; the model construction unit is used to segment the laryngeal image data, extract the tumor volume and laryngeal cavity structure, and construct a three-dimensional anatomical model of the larynx and tumor based on the tumor volume and laryngeal cavity structure; the coordinate system construction unit is used to establish a three-dimensional coordinate system of the three-dimensional anatomical model, and to place a reference magnet A at the anatomical landmark of the larynx, and to mark the position information of the reference magnet A in the three-dimensional coordinate system.

[0085] It is feasible that the deviation correction module includes a real-time measurement unit and a deviation adjustment unit; the real-time measurement unit is used to arrange a magnetic sensor array around the throat, measure the magnetic field generated by the reference magnet A and the guide magnet B, and calculate the three-dimensional coordinates and azimuth of the guide magnet B relative to the reference magnet A by analyzing the magnetic dipole field distribution, thereby obtaining the actual position information of the catheter; the deviation adjustment unit is used to compare the actual position information of the catheter with the target position information corresponding to the optimal path to obtain a deviation vector, and based on the deviation vector, use a PID closed-loop control algorithm to adjust the position and posture of the extracorporeal magnet C to correct the direction of the catheter.

[0086] As an additional implementation method, the system in this embodiment also includes an AI module. The real-time position data of the catheter is provided to the navigation interface for visual display, and abnormal deviation information is provided to the AI ​​module. If an abnormal deviation occurs (for example, it is unable to move forward due to an obstruction), the AI ​​module can determine whether new path planning or manual intervention is required. The introduction of the AI ​​module gets rid of the direct vision dependence of traditional endoscopes and can ensure the accuracy of catheter positioning even in cases of limited field of view. Magnetic positioning + closed-loop control realizes catheter navigation with millimeter-level precision, ensuring that subsequent operations are accurately aimed at the tumor.

[0087] Furthermore, the AI ​​module of this embodiment continuously acquires catheter positioning data and deviation information. If it is detected that the catheter cannot move along the planned path after multiple attempts, the AI ​​will determine that there may be special unmodeled conditions in the laryngeal anatomy (such as abnormal stenosis or tumor obstruction). At this time, the AI ​​can call reinforcement learning or heuristic search algorithms to optimize new path plans online. For example, based on the current catheter position and changes in the surrounding magnetic field, another feasible detour path is inferred and the target waypoint is adjusted. The AI ​​algorithm can fuse multimodal information (magnetic sensing, force feedback, etc.) to make decisions, similar to an autonomous navigation system, and dynamically plan according to environmental changes. At the same time, the AI ​​module can also identify dangerous signals, such as abnormal force on the catheter that may cause entanglement, patient coughing that causes coordinate system drift, etc., and promptly notify the control system to pause or recalibrate.

[0088] The introduction of AI modules imbues the system with brain-like decision-making capabilities, enabling it to assist or automatically make decisions in complex situations. For example, when traditional PID control fails to address complex nonlinear dynamics, the AI ​​controller can approximate nonlinear mappings through neural networks, providing more appropriate control increments. Similarly, in unknown environments, the AI's reinforcement learning strategy can continuously adjust based on rewards (catheter arrival) to improve success rates.

[0089] In a feasible implementation, the modules of the system can form a hierarchical and nested control architecture to improve the modularity and collaborative performance of the system.

[0090] This embodiment adopts a layered task-dependent control architecture, implementing complex directional navigation step by step from the bottom sensor execution layer, the middle task execution control layer, to the top AI supervision decision layer. Each layer collaborates through a clear division of module functions and interface communication to ensure that the system can operate stably and efficiently. Starting from the lower-level modules, the following details the module composition, functional role, information transmission direction, feedback mechanism, and inter-layer dependencies of each layer.

[0091] A. Bottom layer: Sensing execution layer:

[0092] The bottom-level sensing and execution layer interacts directly with the physical world and includes various sensors and actuators. Its main function is to collect environmental and device status data during treatment and execute specific control instructions from the upper layer. The bottom-level module has the smallest scope but the highest operating frequency in this architecture. It is responsible for accurately converting physical quantities into electrical signals in real time and executing corresponding actions, while also feeding back information through interfaces. Typical modules in this layer may include:

[0093] ① Magnetic Positioning Sensor Module: This module detects the position and orientation of the target magnet within the body. This module uses a magnetic sensor array positioned around the patient's throat to sense changes in the external magnetic field signal generated by the magnetic target. It then uses a pre-calibrated model algorithm to calculate the three-dimensional coordinate position and relative orientation of the magnetic target at the tumor site. The magnetic positioning sensor module converts raw sensor signals (such as Hall element voltage) into numerical position data. After preliminary filtering, it is sent to the middle layer in a predetermined format (for example, using an SPI interface for cycle-by-cycle sampling and then sending packaged position data frames via the CAN bus).

[0094] ② Actuator Driver Module: This module is responsible for driving the device's various actuators to achieve physical motion. The positioning mechanism driver submodule controls the mechanical device or electromagnetic coil array used for magnetic navigation positioning. For example, it drives the motor to rotate or linearly move the catheter tip, or adjusts the current output of the magnetic field generator to align the device with the magnetic target in the body. The actuator driver module promptly feeds back execution results and status data to the middle layer, allowing the upper layer to understand the underlying execution status.

[0095] ③ Communication Interface Module (bottom layer): This module implements data transmission and protocol conversion between the bottom layer and the middle layer. Considering the real-time and reliability requirements of medical device control, the bottom layer is typically connected to the middle layer via an industrial fieldbus (such as the CAN bus) or a high-speed serial communication interface. The communication interface module encapsulates data collected by magnetic sensors and actuator status into standardized data frames (for example, using CAN messages with ID and data segments representing various sensor values ​​in a standardized format, or using a custom binary protocol / Protobuf message to define position and dose fields) and continuously sends them to the middle layer at a set update cycle. It also receives control command frames from the middle layer and distributes them to the corresponding driver submodules for execution. This interface module ensures real-time synchronization and integrity verification of data transmission, employing redundancy checks and handshaking mechanisms as necessary to ensure reliable transmission of medical control data.

[0096] B. Middle layer: Task execution control layer:

[0097] The middle layer bridges top-level decisions with underlying hardware, responsible for translating high-level decisions into specific control tasks and coordinating the real-time execution of multiple submodules. It acts as both a "commander" and "coordinator" within the system, achieving precise control of underlying actuators through task scheduling and control algorithms, and performing closed-loop adjustments based on sensor feedback. Middle-layer modules typically include:

[0098] ① Sensor Fusion and State Estimation Module: This module acquires and fuses multi-source sensor data from the underlying layer to provide an optimal estimate of the system state. This module uses algorithms (such as the Kalman filter) to filter and fuse sensor data. For example, it combines the magnetic field position information provided by the magnetic positioning sensor module with the posture data fed back by the actuator to determine the precise position and velocity of the catheter relative to the target. The Kalman filter operates through a prediction-correction loop iteration of the state-space model: the prediction phase predicts the next state based on the device's kinematic model and previous control inputs. The correction phase combines the current measurement values ​​to update the optimal state estimate, effectively reducing noise interference. The output state estimates include the current position, velocity, and navigation error, which are used by other control modules.

[0099] ② Motion control module: responsible for controlling the positioning movement of the catheter so that it is aligned with the magnetic target position of the laryngeal tumor according to the top-level requirements. This module receives the target position or trajectory instruction and calculates the required control quantity based on the current posture provided by the state estimation module. Its control core uses the classic PID algorithm to achieve position closed-loop control: the controller first calculates the position error , for example, the three-dimensional position error vector in millimeters; then according to the PID formula Calculate the control output .in are the proportional, integral and differential coefficients respectively, This can correspond to the motor's drive torque / voltage command or the electromagnetic coil's current. The motion control module performs control calculations at a fixed interval (e.g., every 10ms) and transmits control commands (such as a new motor target position or speed setting) to the underlying actuator driver module via a communication interface. This module's scope covers all positioning-related actuators, ensuring the treatment head accurately tracks the target. Even with slight patient movement or external disturbances, the system maintains alignment through rapid feedback correction.

[0100] ③ Safety Monitoring and Feedback Module: To ensure clinical safety, a safety monitoring mechanism is implemented in the middle layer to continuously monitor system operation. This module monitors the status of each sensor and actuator in real time. If an anomaly is detected (e.g., persistently large positioning error, communication timeout, etc.), it immediately triggers safety measures (e.g., emergency light source shutdown, motor stop, and position lock), and notifies the top-level AI layer and physicians via feedback channels. The safety monitoring module maintains communication with the bottom and top layers by periodically sending health monitoring signals (heartbeats). Redundant communication mechanisms (e.g., retransmitting critical control signals on the CAN bus or reporting emergencies through independent hardware interrupt lines) enhance system reliability. Upon triggering a safety event, the module generates a structured alert message (e.g., in JSON format: { "event": "Safety Alert", "code": 123,"detail": "Light Source Temperature Exceeds Limit"}) and sends it to the top-level AI supervisory decision-making layer. The module can also directly intervene in the control flow (e.g., overriding the motion control module output to zero) to ensure immediate mitigation of the hazard.

[0101] C. Top layer: AI supervision and decision-making layer:

[0102] The top layer is the intelligent decision-making center of the entire system, using artificial intelligence algorithms and global information to provide advanced control and supervision of the navigation process. It makes strategic decisions and adjusts instructions based on feedback data from the middle layer and clinical preset plans. The top layer modules mainly include:

[0103] Continuously monitors the system's operating status and evaluates navigation progress. This module receives aggregated feedback information uploaded by the middle layer (such as task completion status, current positioning error, device operating status, and elapsed time) and compares it with the expected plan to determine whether the strategy needs to be adjusted or the task needs to be terminated. It acts as an intelligent assistant for clinicians. For example, when a large positioning error is detected, the intelligent rollback module is immediately triggered to replan subsequent actions. The global supervision module is also responsible for data logging and traceability management. It continuously records and stores key process data (for example, status logs are recorded every second, with timestamps and saved in JSON line format) for post-analysis and quality control, and presents this information to medical staff through the user interface when necessary. In terms of feedback mechanisms, this module sends detected key events and status information to the AI ​​decision-making module via a dotted arrow path (thus forming a closed-loop adjustment of AI decisions). It also provides feedback to the operating physician in a human-readable format (such as charts and alarm messages) through the human-computer interaction interface module.

[0104] In the three-layer architecture described above, decoupled dependency management is achieved between each layer through clear interfaces and data protocols. The top-level AI supervisory decision-making layer needs to rely on the environmental status and execution results provided by the middle-layer feedback to adjust the high-level strategy; the middle-layer task execution control layer relies on the target tasks issued by the top layer and performs closed-loop control based on the real-time sensor data provided by the bottom layer; the bottom-level sensor execution layer strictly carries out physical actions according to the instructions of the middle layer. Inter-layer communication is coordinated through standard protocols: the top and middle layers use the ZeroMQ message queue mechanism for asynchronous communication, and high-level instructions and status feedback are encapsulated and transmitted in JSON or Protobuf format; the middle and bottom layers synchronously exchange control and sensor data through a real-time bus (such as CAN), and use standardized frame definitions to ensure that different modules have a consistent understanding of the information. In terms of data structure, the information exchanged by each layer is represented in the form of key-value pairs or binary fields. For example, a positioning command message sent by the top layer might include the field set { "target_pos": [x,y,z], "tolerance": ε} , while the dose status feedback from the bottom layer might include a field structure such as { "dose_delivered": D_{current}, "dose_target": D_{target}} . This unified message format ensures clear and reliable task coordination logic between different modules. Regarding task coordination, each treatment task defined by the top layer defines expected conditions and feedback metrics to be monitored. The middle layer uses this to schedule the execution of subtasks and perform condition checks, while the bottom layer provides real-time feedback during execution, forming a closed loop. This inter-layer dependency and coordination mechanism ensures that the entire treatment process proceeds as expected and is capable of dynamic adjustments based on feedback.

[0105] Example 2

[0106] This embodiment is further described below.

[0107] Medical imaging data (such as CT or MRI) from the patient's surgical area is acquired to reconstruct a 3D anatomical model and mark the tumor location, providing a precise virtual model for navigation. The magnetic positioning navigation subsystem consists of a surface positioning magnet A, a guide magnet B at the front end of the catheter, a movable external strong magnetic traction device (external magnet C), a magnetic sensor array, and a controller, used to locate and control the catheter's position in real time.

[0108] Surface positioning magnet A is fixedly placed at a specific location on the patient's body surface (such as the fourth cervical vertebra in the back of the neck) and serves as the origin of the navigation coordinate system. A guide magnet B is installed at the front end of the photodynamic catheter, which is inserted along the patient's throat or esophagus to the area near the tumor. The external magnet C consists of a high-strength permanent magnet mounted on a mechanically controllable three-dimensional mobile platform (or robotic arm) that can move and orient within a specific range outside the patient's body. Magnetic sensor arrays (such as multi-directional Hall sensors or rotating permanent magnet positioning systems) are arranged around the operating table to detect the spatial position and posture of magnets A, B, and C in real time. The controller calculates the position coordinates of the guide magnet B based on the data collected by the magnetic sensors and displays them in real time in the navigation model.

[0109] The photodynamic catheter consists of a flexible catheter body and an internal optical fiber or LED light source. In addition to the guide magnet B, its front end is also equipped with an optical component (such as a lens or reflective cone) for directional light emission, forming a directional beam to enhance the irradiation intensity of the tumor target area. The catheter is connected to the external light source / laser through a controllable delivery mechanism and can be mechanically propelled under the traction of the magnet to achieve the forward, backward and steering of the catheter within the body. The entire system also includes a computer console and a surgical navigation display module, which presents real-time information such as the patient's anatomical model, tumor location, catheter trajectory, and magnet position for the surgeon's reference and monitoring.

[0110] The catheter moves along a predetermined path toward the tumor target area, continuously monitoring the catheter's position and posture and comparing it with the tumor position in the virtual model. If the catheter deviates from the planned path or the patient moves, the system will adjust the position and posture of the external magnet C through feedback, and control the catheter propulsion mechanism if necessary, to correct the catheter trajectory in real time. Through this closed-loop control, dynamic navigation of the catheter to the tumor surface is achieved. The entire process does not require endoscopic imaging. Magnetic navigation ensures that even if the tumor causes stenosis and the endoscope cannot be inserted, the catheter can still reach the target area smoothly.

[0111] For larger tumors, this process can be repeated for multiple rounds of navigation.

[0112] As an optional feature, this embodiment also includes a dose calculation module. Based on parameters such as tumor size, depth, and tissue optical properties provided by a preoperative 3D model of the patient's larynx, this module utilizes a deep learning model to calculate the laser power and irradiation duration required to achieve the desired therapeutic dose. This module utilizes models of intra-tissue light transmission (such as the Beer-Lambert theorem or Monte Carlo simulation) to quantitatively estimate the surface incident light intensity required to achieve an effective light dose within the tumor, thereby determining the light source output setting and exposure time. For example, based on tumor thickness, the module can calculate the required intensity and duration at the tumor surface to ensure sufficient light dose to deeper tissues. Furthermore, the aforementioned magnetic navigation technology in this system enables the tip of the light guide to precisely conform to the tumor surface, thereby improving light energy coupling efficiency. This conformation process, due to the close contact between the light guide tip and the tumor surface, can cause a certain degree of deformation in the local skin or soft tissue. The dose calculation module also factors this tissue deformation into its modeling and modifies the geometric parameters of the 3D model accordingly to improve the accuracy of dose estimation and ensure the calculated treatment dose is reliable and effective.

[0113] As an add-on implementation method, this embodiment also includes a light source and fiber calculation module: this module recommends specific laser light source power settings and fiber output parameters based on the dose calculation results. In preoperative planning, it determines the optimal positioning and distance of the fiber tip relative to the tumor surface. For example, the planned fiber end should be several millimeters away from the tumor surface to form a laser spot of appropriate size to cover the entire tumor area, ensuring that the light beam is effectively focused on the tumor tissue and reducing energy loss. Through the parameter recommendations of this module, the required laser wavelength, power, fiber layout and other factors are clearly defined in the surgical plan, providing a basis for subsequent implementation.

[0114] As an optional feature, this embodiment also includes an irradiation planning module. When a tumor is large or irregularly shaped, this module automatically plans multiple irradiation points and their corresponding sequence based on the preoperative model's determination that a single irradiation point is insufficient to cover the entire lesion. It outputs multiple optimized illumination target locations and an irradiation sequence to ensure that all tumor components receive adequate radiation doses during subsequent treatment. This planning output guides the navigation module to position the light guide to each target point during surgery, achieving a zoned, multi-point irradiation plan and preventing under-irradiation of any tumor region.

[0115] The final illumination plan contains detailed parameters such as the required laser power, exposure time, and irradiation position, and is combined with the navigation and positioning process to form a complete preoperative treatment planning program, providing doctors with a specific reference plan for related photodynamic therapy illumination, but it is not directly used as a treatment plan.

[0116] Figure 1This is a flowchart of the overall system of the present invention: the process shown covers the entire process from preoperative CT image input, path planning, and magnetic navigation control. The process first obtains the patient's CT / MRI and other medical imaging data and reconstructs a three-dimensional model of the larynx. The tumor location and surrounding anatomical structures are extracted as input for path planning. The path planning module then calculates the optimal catheter path from the patient's natural cavity to the tumor based on the model and transmits the result to the navigation control module. The control system then controls the external magnetic field to guide the catheter movement according to the planned path, adjusting the magnet position and posture in real time to ensure that the catheter tip reaches the target tumor along the predetermined trajectory.

[0117] Figure 2 This diagram illustrates the system architecture and preoperative modeling: It depicts the overall framework for reconstructing a laryngeal tumor model based on patient images and the layout of the magnetic navigation system. The diagram includes the patient's laryngeal anatomical model (with the tumor location marked), the external magnetic navigation device (magnets on the robotic arm), and the control / computing unit. Arrows indicate the preoperative data flow: CT / MRI images are processed by AI to generate a 3D model, and the magnetic field distribution is simulated for planning purposes.

[0118] Figure 3 This is a flowchart of the underlying path planning algorithm of the present invention: the process detailedly illustrates the steps of path point generation, constraint processing, cost function calculation, and optimization solution. The algorithm first generates an initial path point sequence based on the preoperatively acquired three-dimensional anatomical model of the laryngeal region and the target tumor location. The algorithm then applies constraints such as anatomical pathway diameter restrictions, avoidance of critical tissue structures, and maximum catheter bend radius to the path, adjusting it to ensure that each path segment conforms to physical and anatomical constraints. A cost function is then constructed to reflect the quality of the path. For example, factors such as path length, curvature smoothness, and obstacle avoidance distance are weighted and integrated to evaluate the comprehensive cost of different candidate paths. Finally, an optimization algorithm is used to iteratively solve for the optimal path, minimizing the cost function while satisfying all constraints. The optimized optimal path is output as a series of discrete path nodes, which are subsequently called upon by the navigation control module to ensure the catheter follows this path.

[0119] Figure 4 This is a schematic diagram of the magnetic navigation control principle: It illustrates the operation of an extracorporeal magnetic field-guided catheter. In the cross-sectional view of the patient's head and neck, the catheter enters the throat through the mouth. The tip of the catheter, with a magnet (a small black dot), is attracted by an external magnet. The magnet is mounted at the end of a robotic arm and can be moved to the outside of the patient's neck. The dotted line represents the planned path, along which the catheter moves. The adjacent control block diagram illustrates the closed loop of magnetic navigation: position sensor → controller → magnetic actuator → catheter movement → re-sensing.

[0120] Experimental comparison:

[0121] Table 1 below compares the performance of different path planning algorithms. The table compares the planning time and path accuracy of the RRT algorithm, the A* algorithm, and the method of the present invention. This comparative experiment was conducted under the same laryngeal navigation scenario. Each algorithm calculated the catheter path and output the corresponding results. Planning time and path quality indicators (such as path length and obstacle avoidance margin) were then calculated. The table shows that the path planning algorithm of the present invention generates high-quality paths in a shorter time. In comparison, the traditional RRT algorithm, due to its random expansion strategy, plans longer paths and takes a moderate amount of time. The A* algorithm can find a relatively good path, but takes a long time due to its large search state space. The algorithm of the present invention, however, significantly reduces computational overhead through a specific optimization strategy. It outperforms other algorithms in terms of ensuring short path length, smooth turns, and sufficient safety distance. Planning time is also significantly reduced, demonstrating its combined advantages in efficiency and accuracy.

[0122] Table 1

[0123]

[0124] As shown in the table, the AI ​​intelligent planning algorithm of our present invention demonstrates comprehensive advantages across various metrics. First, in terms of efficiency, our method takes an average of only 0.7 seconds to plan, significantly shorter than the 1.8 seconds of RRT and 1.2 seconds of A. This is because the DRL strategy outputs near-optimal paths in real time, avoiding the inefficiency of random sampling in RRT and the redundant grid search required by A. Second, in terms of path quality, our method plans the shortest average path length (approximately 118.5 mm), which is over 13% shorter than RRT. Its path smoothness score is near full, with a maximum curvature of only 0.83 rad / m, indicating smoother, more gradual turns and less abrupt turns. In contrast, RRT paths are tortuous (maximum curvature 1.45 rad / m) and have a redundancy rate of approximately 14%, while A paths have a redundancy rate of approximately 7%. Our method, however, achieves near-zero redundancy (equal to the theoretical optimal path), achieving near-shortest path planning. Furthermore, our method also offers superior safety: the minimum safety margin for the path from anatomical obstacles is 6.2 mm, significantly greater than that of RRT and A (approximately 2–3 mm, respectively). This demonstrates that AI planning is superior at navigating confined spaces and avoiding close proximity to normal tissue. In terms of navigation accuracy, our method accurately guides the catheter to its target, with a terminal positioning error of only approximately 0.2 mm. In comparison, RRT, due to randomness, sometimes has an endpoint deviation of approximately 1–2 mm, while A, limited by grid resolution, has an endpoint error of approximately 1 mm. In terms of trajectory repeatability, the paths of A and our method are highly consistent, with nearly identical planning results each time (score = 5), while the RRT algorithm suffers from high path variability due to random sampling (repeatability score of only 2). Finally, in terms of computational complexity, A* requires traversing a large number of nodes (high complexity), while RRT has limited convergence speed in high-dimensional spaces (medium complexity). However, our algorithm only requires forward network inference, resulting in high real-time planning and low computational overhead. Overall, our AI path planning algorithm significantly outperforms traditional algorithms in navigation accuracy, path smoothness, and planning efficiency. This result strongly demonstrates the feasibility and superiority of introducing deep reinforcement learning into minimally invasive navigation path planning.

[0125] Table 2 below is a comparison of the catheter path tracking error before and after the introduction of feedback control: The table shows a comparison of the changes in navigation accuracy before and after the closed-loop control strategy intervenes. The horizontal axis is time, and the vertical axis is the deviation distance of the catheter front end relative to the planned trajectory. When feedback control is not adopted (open-loop control), the catheter path tracking error is large and decays slowly, and the curve shows a significant deviation for a long time; after adding closed-loop feedback control, the system continuously adjusts the movement of the magnet according to the real-time position deviation, and the error curve drops rapidly and converges to a position close to zero, allowing the catheter to move more accurately along the planned trajectory. This comparison shows that the introduction of real-time imaging / sensing feedback for closed-loop control significantly improves the accuracy and stability of the navigation process, reduces steady-state errors, and accelerates the convergence speed of catheter positioning.

[0126] Table 2

[0127]

[0128] As shown in the table, the introduction of real-time feedback control significantly improves catheter tracking accuracy and response speed. As shown in Table 2, closed-loop control reduces the catheter's steady-state error from 2.7 mm in the open-loop control to only 0.6 mm, improving steady-state accuracy by approximately 4.5 times. The instantaneous maximum deviation also decreases from 6.3 mm to 2.1 mm, indicating a two-thirds reduction in the catheter's peak deviation from the planned trajectory. Furthermore, the closed-loop system achieves 90% error convergence time of only 1.2 seconds, over 65% shorter than the open-loop control's 3.5 seconds, demonstrating that the addition of feedback allows the catheter to more quickly converge to the planned trajectory. Dynamically, the error curve decays slowly under open-loop control, leaving a significant residual error. However, closed-loop control significantly accelerates error decay through continuous correction. Data for the integrated error (the area under the error curve between 0 and 5 seconds) show that the closed-loop system is only approximately 5 mm·s, significantly lower than the open-loop system's 18 mm·s, reducing the cumulative error by over 70%. Furthermore, the closed-loop control was properly adjusted, resulting in no significant overshoot (no rebound overshoot in the error). While the open-loop control lacked correction, there was no overshoot but a significant steady-state deviation persisted. Overall, the introduction of real-time position correction in feedback control effectively suppressed navigation errors (error suppression ratio ≈ 4.5), enabling the catheter to more accurately and rapidly follow the planned path. This result demonstrates the significant role of the closed-loop magnetic navigation control strategy employed in this invention in improving catheter positioning stability and accuracy.

[0129] Table 3 below compares the time taken for magnetic navigation and traditional manual manipulation in a clinical model. This table compares the time taken to complete tumor localization using magnetic navigation with the device of the present invention and traditional manual catheter manipulation. The horizontal axis represents the catheter positioning method (automated magnetic navigation vs. manual manipulation), and the vertical axis represents the average time required to position the catheter tip at the target tumor. This comparison of the two methods demonstrates that the magnetic navigation system of the present invention, which automatically performs path planning and guidance, significantly reduces catheter positioning time. The system rapidly delivers the catheter to the target according to the pre-planned path, typically completing positioning in a shorter time. In contrast, manually advancing and adjusting the catheter through the narrow and tortuous laryngeal passage in the clinical model often takes significantly longer. This comparison demonstrates the advantages of the device of the present invention in surgical efficiency, potentially reducing surgical duration and anesthesia burden on patients, and improving the safety and efficiency of clinical procedures.

[0130] Table 3

[0131]

[0132] As shown in the table, automated magnetic navigation significantly outperforms manual positioning in terms of efficiency and consistency. This system does not rely on operator skill; even ordinary clinicians can initiate automated positioning with a single click. In contrast, the manual method requires experienced clinicians to manually adjust the catheter position by inserting the endoscope. As shown in Table 3, magnetic navigation completes catheter positioning in an average of approximately 7.8 seconds, while manual operation takes an average of nearly 18 seconds, a time reduction of over 50%. Even accounting for inter-procedural variability, automated navigation time remains stable within a range of 6 to 9 seconds, with a standard deviation of only 1.2 seconds. Manual operation time exhibits significant variability, with a maximum of 24.3 seconds and a wide range of fluctuations (approximately 12 to 24 seconds), reflecting the significant influence of individual operator rhythm and significant variability between attempts. Due to its precise and repeatable calculations, the automated system achieves nearly identical path planning and execution, with a positioning error of only 0.3 mm across repeated trials, demonstrating high repeatability. In contrast, manual operation can exhibit slight deviations in the endpoint position between attempts, with a repeatability error of approximately 1 mm. In terms of positioning accuracy, automated navigation relies on imaging feedback and magnetic targeting fine-tuning, precisely aligning the catheter tip with the tumor (error <0.5 mm). Manual methods, relying primarily on visual estimation and feel, often have an error of 1–2 mm when aiming the catheter at the target. This demonstrates that the magnetic navigation system of the present invention significantly improves the speed and precision stability of catheter positioning. Faster positioning helps shorten patient anesthesia time and improve surgical efficiency; greater consistency reduces the need for repeated adjustments, enhancing clinical safety.

[0133] Table 4 below is a table comparing the navigation success rates under different lesion locations or anatomical pathways: The table shows a comparison of the probability of the system of the present invention successfully navigating the catheter to the target under various tumor locations and approach conditions. The horizontal axis represents different scenario conditions (for example, the tumor is located in the epiglottis area, the vocal cord area, the subglottis area, or the catheter enters through the oral cavity or through the nasal cavity), and the vertical axis represents the percentage success rate of the catheter successfully reaching the tumor target area under the corresponding conditions. Each bar represents the statistical success rate of multiple tests under that condition. As can be seen from the table, the system of the present invention exhibits a high navigation success rate in various anatomical scenarios: in tumor sites with unobstructed pathways, the navigation success rate is close to 100%; even in cases where the anatomical space is small and the pathway is complex, the system's navigation success rate remains at a high level (for example, there are only a few failure cases), fully demonstrating the adaptability of the device of the present invention to different laryngeal tumor locations and the robustness of the navigation scheme.

[0134] Table 4

[0135]

[0136] As shown in the table, the data from different anatomical scenarios show that the system of the present invention has good adaptability and robustness. For relatively easy epiglottal region tumors, the system can complete navigation positioning with a success rate of nearly 100% regardless of the oral or nasal route. The path complexity of this location is low (score 23), and there is almost no obstacle for the catheter to reach the lesion directly, so there are no failure cases, which reflects the reliability of the system under a smooth path. Because the glottic region tumor is located in the middle part of the larynx, the path needs to bypass the epiglottis to enter the glottis, and the complexity is medium (56); the data show that the system success rate is still above 90%. A few failures mainly occurred in the nasal approach (success rate 90%), because the catheter positioning was slightly difficult when passing through the narrow part of the vocal cord, but overall, most of the success rates were achieved regardless of the oral or nasal approach. For the most challenging deep subglottic tumors, the path complexity score was as high as 89, and it was necessary to pass through the narrow subglottic cavity, and magnetic navigation was the most difficult; despite this, the system success rate reached 90% for the oral approach and 85% for the nasal approach. The reasons for failure are mostly related to deep navigation: first, magnetic attenuation - as the catheter goes deeper into the larynx, the traction force generated by the external magnetic field at the target weakens, which may cause the catheter to fail to reach the final position in individual attempts; second, path resistance - the narrow and curved channel increases the friction between the catheter and the tissue, and in some cases the advancement of the catheter is hindered. However, these failures are only a minority, and the overall success rate remains at a high level, indicating that the navigation algorithm of the present invention has good robustness and adaptability to different anatomical structures. Additional minimally invasive and visual field scores show that: all approach plans are operated through the natural cavity, the minimally invasive scores are all 45, and no additional tissue incision is caused; the transoral route requires the use of an opener or laryngoscope to expose the glottis, which slightly increases the burden on the patient during deep operations (the score is slightly lower than the transnasal route), and the transnasal route uses fiber endoscope navigation, which is more flexible and comfortable for patients. Regarding surgical field visualization, the epiglottis, due to its superficial location, is clearly visible both directly and endoscopically (score 45). However, the glottis and subglottis are partially obscured by the epiglottis or vocal cords, limiting direct visualization. Transnasal endoscopy allows for access to the lesion, but with a narrower field of view (score 34). Despite these differences in field of view, the system achieved a high success rate of automated navigation under various path conditions, reducing reliance on manual visual linear vision. This demonstrates the effectiveness of AI magnetic navigation in complex anatomical environments and holds the potential to improve access to treatment in previously inaccessible areas.

[0137] Table 5 below shows the catheter response time under different magnet control strategies: The curves shown reflect the differences in the response speed of the catheter front end to the target instructions under each strategy. The horizontal axis is time, and the vertical axis is the position deviation or displacement of the catheter front end. Curve A represents the response when the permanent magnet is moved mechanically for guidance. It is characterized by a slow initial response and a significant lag. Curve B represents the response when the electromagnetic coil is used to quickly change the magnetic field for guidance. Its rising section is steeper, and the catheter front end can approach the target position more quickly with a smaller steady-state error. As can be seen from the table, there are significant differences in the catheter response speed caused by different control strategies. Among them, the strategy based on electromagnetic rapid control can achieve faster response and higher steady-state accuracy, improving the control performance of the system during dynamic navigation.

[0138] Table 5

[0139]

[0140] As shown in the table, the magnet control strategy has a significant impact on the response characteristics of the catheter. As shown in Table 5, the electromagnetic coil fast field control scheme is significantly better than the mechanical moving permanent magnet scheme in dynamic response. The initial response delay of the catheter during electromagnetic control is only about 0.12 s, and the magnetic field change can be generated almost instantaneously, and the front end of the catheter begins to move quickly; while the mechanical moving permanent magnet is affected by the start and stop of the motor and inertia, the catheter response lags by about 0.30 s. In terms of the time required to reach half the distance to the target position, the electromagnetic scheme completes half of the positioning process in about 0.8 s, which is nearly twice as fast as mechanical movement (1.5 s). This is also reflected in the acceleration of the catheter movement: the electromagnetic coil can provide a large magnetic gradient in an instant, driving the catheter at a speed of about 60 mm / s 2Starting from a maximum acceleration of approximately four times that of the mechanical approach, the electromagnetic approach achieves a steeper upward trajectory (faster response). Thanks to its faster dynamic response, the electromagnetic approach achieves a steady-state error of only 0.4 mm, significantly less than the 1.0 mm of the mechanical approach. This faster and more precise control ensures the catheter's final position is closer to the target. On the other hand, the two approaches have their own unique characteristics in terms of resource consumption and implementation complexity. The control algorithm for the manipulator's permanent magnet movement is based on traditional robot kinematics, offering low computational overhead and a control frequency typically of several dozen times per second (due to the slow physical response of the mechanical system, a feedback frequency of approximately 50 Hz is sufficient). Its energy consumption primarily comes from the power consumption of the drive motor to move the magnet. This experiment estimated the energy consumption for a single positioning operation to be approximately 50 J. In contrast, the electromagnetic coil approach requires real-time calculation of the coil current to generate the required magnetic field vector. This complexity, especially when using a multi-coil control scheme, results in a more complex algorithm and control logic (rated as "high" in cost). However, the electromagnetic approach offers a faster response and supports higher control update frequencies (theoretically, up to hundreds of hertz). The actual feedback control frequency in this system is approximately 200 Hz, resulting in more continuous and smooth magnetic field regulation. This high-speed closed-loop control further enhances the fine-tuning capabilities of catheter navigation. Regarding energy consumption, the electromagnetic coil consumes electrical energy to generate the magnetic field. Although a single rapid movement takes a short time (<1 s), continuous power may be required to maintain catheter stability. The estimated energy consumption is approximately 120 J, slightly higher than that of the mechanical movement solution. However, at an acceptable increase in energy consumption, the electromagnetic solution offers significant performance improvements—shorter response delay and smaller positioning error. Experimental curves show that the catheter response curve is initially flat with significant lag when the permanent magnet is mechanically moved; in contrast, the curve rises steeply when the electromagnetic coil is controlled, allowing the catheter to rapidly approach the target. In summary, while the electromagnetic dynamic field control strategy has slightly higher implementation complexity and power consumption, it significantly improves the dynamic performance and accuracy of the navigation system, demonstrating the innovative design of the magnetic control module in this invention: prioritizing real-time navigation and accuracy within acceptable resource constraints, tailored to surgical needs.

[0141] Table 6 below provides a comprehensive comparison of the navigation system's intelligence, robustness, efficiency, and safety. It comprehensively quantifies the differences between the proposed magnetic-targeted AI navigation system and traditional manual operation in terms of intelligence, system robustness, treatment efficiency, and safety. AI intelligence is measured by DRL training duration and decision frequency; robustness considers sensor noise sensitivity and error stability; efficiency is measured by the number of successful positioning attempts per unit time; and safety compares tissue invasion and magnetic field safety.

[0142] Table 6

[0143]

[0144] As shown in the table, a comprehensive comparison demonstrates that the magnetic navigation photodynamic therapy device of the present invention offers outstanding advantages in terms of intelligence, robustness, efficiency, and safety. First, regarding intelligence, the system incorporates AI algorithms such as deep reinforcement learning as its core decision-making framework. While extensive initial simulation training (approximately dozens of hours) is required to learn optimal strategies, once training is complete, the system can output approximately 10 decision commands per second during surgery, adjusting catheter and illumination parameters in real time. In contrast, traditional manual control relies entirely on surgeon experience, lacks adaptive intelligence, and has a limited human response rate, making only one or two adjustments per second. As has been noted, deep learning control surpasses traditional algorithms in trajectory smoothness and adaptability. Second, regarding robustness, the AI ​​navigation system is insensitive to sensor noise and environmental interference: moderate positioning errors or image noise do not significantly affect control decisions, and the algorithm maintains stability (high error rejection ratio) through feedback correction. Manual operation, on the other hand, requires a clear field of view, and obstructions to the endoscopic view (e.g., bleeding or fog) can lead to misjudgment or positioning errors. Repeated testing has shown that the standard deviation of the final positioning error of this system is less than 0.2 mm, demonstrating highly stable positioning accuracy. In contrast, manual positioning is subject to subjective factors and has limited error stability. Third, in terms of treatment efficiency, the magnetic navigation system can automatically optimize the path, achieving a single-pass catheter positioning success rate of at least 95% even in complex anatomical conditions, with an average time of only 7.8 seconds, significantly exceeding the success rate and time efficiency of manual methods. Converted to a unit of time, the machine can complete approximately 7-8 precise positioning attempts per minute, while manual methods can only achieve approximately 3 attempts. This higher efficiency means reduced surgical time and anesthesia risks. Finally, in terms of safety, both methods are minimally invasive through natural cavities, causing no additional trauma to the patient (no tissue invasion). However, traditional surgical procedures often require surgical intervention when positioning is difficult, which significantly increases the trauma and risk. This system utilizes magnetic field remote control to reach deep lesions without surgery, achieving a zero-invasive treatment option. In addition, the magnetic field strength and change rate used in the present invention are within a safe range, no significant increase in tissue temperature (<1°C) was detected, and there is no risk of magnetic field thermal effect; in comparison, although traditional operations have no magnetic field concerns, there may be a risk of complications due to longer operation times in complex operations.

[0145] Example 3

[0146] This embodiment further provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0147] Example 4

[0148] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0149] Example 5

[0150] This embodiment also provides a computer program product, including a computer program, which implements the steps of the method when executed by a processor.

[0151] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An in vitro magnetic target positioning and navigation system for the larynx, characterized in that: include: A model building module, a path planning module, a magnetic control guidance module, a deviation correction module, a safety monitoring module and an intelligent rollback module are connected in sequence; a model construction module, configured to construct a three-dimensional anatomical model of the larynx and target points based on the laryngeal image data, and mark the position information of a reference magnet in the three-dimensional anatomical model; wherein the reference magnet is placed at an anatomical landmark of the larynx; A path planning module is used to plan an optimal path for the catheter from the entrance to the target point in the larynx in a three-dimensional anatomical model based on a path planning algorithm; The magnetic control guidance module is used to embed a guide magnet at the front end of the catheter, move the external magnet according to the optimal path, and perform magnetic traction and directional control on the catheter based on the magnetic force and magnetic dipole torque generated by the external magnet acting on the guide magnet; A deviation correction module is used to track the position of the front end of the catheter in real time using a magnetic sensor array and perform deviation correction of the travel path through closed-loop control; The safety monitoring module is used to obtain the distance information between the catheter and the surrounding tissue in real time. When the real-time distance is less than the preset threshold, a contact warning is issued; An intelligent rollback module is used to pause the current navigation path and re-plan the catheter's path based on collision warnings to ensure that the catheter reaches the target location; The deviation correction module includes a real-time measurement unit and a deviation adjustment unit; The real-time measurement unit is used to arrange a magnetic sensor array around the throat, measure the magnetic field generated by the reference magnet and the guide magnet, and calculate the three-dimensional coordinates and azimuth of the guide magnet relative to the reference magnet in real time by analyzing the magnetic dipole field distribution, thereby obtaining the actual position information of the catheter; The deviation adjustment unit is used to compare the actual position information of the catheter with the target position information corresponding to the optimal path to obtain a deviation vector. Based on the deviation vector, a PID closed-loop control algorithm is used to adjust the position and posture of the extracorporeal magnet to correct the direction of the catheter.

2. The system according to claim 1, wherein: It also includes a dose calculation module, a light source and optical fiber calculation module, and an irradiation planning module; A dose calculation module, which uses a deep learning model to obtain the optimal dose based on a 3D anatomical model of the larynx and target. a light source and optical fiber calculation module, connected to the dose calculation module, for determining the laser light source power setting and optical fiber output parameters based on the optimal dose; The irradiation planning module is connected to the light source and fiber calculation module and the intelligent rollback module respectively, and is used to optimize the irradiation target position and irradiation sequence. When the catheter reaches the target target position, the target target is irradiated in combination with the laser light source power setting and the fiber output parameters.

3. The system according to claim 1, wherein: The model building module includes: a model building unit and a coordinate system establishment unit; The model building unit is used to segment the laryngeal image data, extract the target target volume and laryngeal cavity structure, and build a three-dimensional anatomical model of the larynx and the target target based on the target target volume and laryngeal cavity structure; The coordinate system construction unit is used to establish a three-dimensional coordinate system of the three-dimensional anatomical model, and to place a reference magnet at an anatomical landmark of the larynx and mark position information of the reference magnet in the three-dimensional coordinate system.

Images