A non-structured visual guidance based plasma adaptive scanning treatment system and method

Abstract

This invention discloses a plasma adaptive scanning therapy system and method based on unstructured visual guidance, belonging to the interdisciplinary field of high-end medical equipment and intelligent robotics. The system includes a multimodal visual perception subsystem, an intelligent decision-making and planning center, a multi-axis flexible actuator, and an active safety monitoring module. The core of this invention lies in utilizing machine vision to directly identify unstructured hand-drawn markings on the patient's body surface by a doctor, converting the image coordinates into the physical coordinates of the robotic arm through hand-eye coordinate mapping. The intelligent decision-making center, based on a "dose-motion coupling model," dynamically inversely solves the target treatment dose into the optimal movement speed along the scanning path, achieving an intuitive interaction from "doctor drawing circles" to "automatic and precise machine scanning." The system employs a dual active safety mechanism of "global visual monitoring + local laser closed-loop": high-frequency global visual comparison of body surface displacement and real-time compensation of Z-axis height via local laser ranging, ensuring treatment safety even under respiratory fluctuations or micro-movements. This invention solves the technical challenges of uneven dosage and low efficiency in traditional handheld treatments, as well as the inability of traditional industrial automation equipment to adapt to "non-standard markings" and "flexible dynamic targets" in medical scenarios, achieving standardized, intelligent, and precise treatment of large-area, irregular body surface lesions.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of high-end medical equipment and intelligent robots, specifically relating to a treatment system and method that utilizes machine vision to recognize doctors' hand-drawn markings and achieves adaptive plasma plume path planning and safe scanning through a multi-axis motion platform. This technology is particularly suitable for dermatology, burn surgery, and oncology, for precise plasma treatment of large-area, irregular lesions on the body surface. Background Technology

[0002] Cold atmospheric plasma (CAP), as an emerging physical therapy method, has shown great potential in clinical applications due to its excellent bactericidal, healing-promoting, and tumor cell apoptosis-inducing properties. However, existing plasma therapy equipment generally suffers from the dual bottlenecks of "low efficiency in human-machine collaboration" and "difficulty in treatment standardization," specifically manifested in the following ways: 1. Limitations of manual operation: Existing equipment mostly uses a handheld pen-shaped handle, and the output plasma plume is a thin, long beam (usually 2mm in diameter). When treating large-area burns or skin lesions, doctors need to perform reciprocating sweeping motions like "spraying paint." This operation is highly dependent on the doctor's hand stability and experience, and is prone to uneven distribution of treatment dose (energy density) due to hand tremors and uneven movement speed, resulting in "missed" or "overtreated" areas, seriously affecting the consistency of treatment efficacy.

[0003] 2. The "Incompatibility" of Industrial Automation: While mature robotic arm painting and welding technologies exist in the industrial sector, their core logic is based on pre-set rigid CAD models and fixed workpiece positions. Directly transplanting them to medical scenarios faces three major challenges: (1) Difficulty in converting intentions: Doctors cannot draw precise CAD drawings in emergency treatment, and industrial equipment cannot understand the doctor's intuitive "hand-drawn circle" intention.

[0004] (2) Poor environmental adaptability: The human body surface is flexible, curved and moves slightly with breathing (especially the treatment sites in the chest and abdomen). Fixed industrial paths cannot adapt to this dynamic change and there is a risk of collision.

[0005] (3) Difficulty in parameter decoupling: Industrial control usually sets the path and process parameters (such as power and speed) separately, while in medical treatment, the scanning speed directly determines the biological dose, and the two must be dynamically coupled.

[0006] Therefore, there is an urgent need for an intelligent plasma therapy system that can directly understand doctors' unstructured clinical intentions (such as marking), adapt to the flexible environment of the human body, and achieve precise and controllable dosage. Summary of the Invention

[0007] I. Technical Objectives This invention aims to solve the technical challenges of low efficiency and uneven dosage in manual treatment, as well as the inability of traditional industrial automation equipment to adapt to "non-standard markings" and "flexible dynamic targets" in medical scenarios. The core lies in constructing a "what you see is what you get" visual interaction logic and a dynamic closed-loop control system of "perception-decision-execution".

[0008] II. Technical Solution

[0009] To achieve the above objectives, this invention provides a plasma adaptive scanning therapy system and method based on unstructured visual guidance. This system constructs a closed-loop system encompassing "visual perception - intelligent decision-making - flexible execution - proactive safety," transforming the doctor's unstructured hand-drawn intentions into high-precision automated treatment actions.

[0010] (I) System Architecture Scheme The system provided by this invention mainly includes the following four core subsystems: 1. Multimodal visual perception subsystem The multimodal visual perception subsystem includes a high-resolution global shutter industrial camera, a specific band spectral enhancement light source, and a laser rangefinder.

[0011] The multimodal visual perception subsystem is responsible for high-fidelity image acquisition of the treatment area. It enhances the contrast between the doctor's hand-drawn markings and the skin using a specific wavelength light source, acquires images using a high-resolution camera, and, in conjunction with a laser rangefinder, monitors the micro-distance between the treatment head and the body surface in real time, providing the system with precise visual and distance data input.

[0012] The functional logic includes: a camera vertically mounted above the treatment area to capture high frame rate images containing hand-drawn markings by the doctor; a spectrally enhanced light source using a wavelength complementary to the color of the medical marker (e.g., a yellow / white high color rendering index light source for blue ink) to maximize the contrast between the markings and the skin background; and a laser rangefinder integrated into the output port for real-time measurement of the micro-distance between the treatment head and the patient's body surface.

[0013] To further enhance robustness in complex clinical environments, the multimodal visual perception subsystem also integrates a deep learning semantic segmentation engine. This engine employs a lightweight U-Net network architecture, and its pre-trained label segmentation model can perform confidence verification and correction on Hue Saturation Value (HSV) threshold segmentation results, making it particularly suitable for interference scenarios such as scars, pigmentation, and ointment residue. The system integrates color features, morphological features (line continuity, curvature changes), and depth segmentation results through a voting mechanism to output high-confidence candidate label regions. Simultaneously, the human-computer interaction terminal provides a "manual correction" mode, allowing doctors to directly add or erase recognition areas on the touchscreen, ensuring workflow continuity under extreme lighting or complex backgrounds.

[0014] 2. Intelligent Decision-Making and Planning Center The intelligent decision-making and planning hub includes: a built-in image semantic segmentation engine, a hand-eye coordinate mapping module, a dose-motion coupling solver, and a dynamic path planner.

[0015] The intelligent decision-making and planning center, acting as the system's "brain," is responsible for transforming unstructured hand-drawn intentions into machine instructions. It identifies and repairs hand-drawn outlines using image algorithms, mapping them to three-dimensional physical coordinates. Simultaneously, based on a dose-motion coupling model, it dynamically plans a full-coverage scanning path and calculates the optimal motion speed to ensure precise and controllable treatment dosage.

[0016] The central processing logic for intelligent decision-making and planning includes: Unstructured intent recognition: An adaptive threshold segmentation algorithm based on the HSV color space is used, combined with morphological operations (closing operation) to repair the breaks in the hand-drawn lines and extract the mask of the closed treatment area.

[0017] Hand-eye coordinate mapping: Using a pre-calibrated hand-eye transformation matrix, irregular contours in the image pixel coordinate system are converted in real time into a set of three-dimensional physical boundary points in the robot arm base coordinate system.

[0018] The intelligent decision-making and planning center includes a dose-motion coupling solver, which is based on a treatment dose model. Dose-coupled solution: Establishing a treatment dose model

[0019] Where: D is the therapeutic dose (J / cm²) 2 P is the power (W), η is the energy coupling efficiency, v is the scanning speed (cm / s), and w is the effective width of the plume (mm).

[0020] The dose-motion coupling solver dynamically inversely solves the theoretical scanning speed v of each path point based on the set target dose, current power, and effective plume width.

[0021] When the plasma therapy head uses two or more output ports to output simultaneously, the effective plume width w is the sum of the effective plume widths of the several output ports. For example, if there are three output ports, and the effective plume width of each output port is 2mm, then the effective plume width is 6mm.

[0022] To accommodate tissue heterogeneity and dynamic biological effects, the dose coupling solver supports sub-region division of the treatment area and differentiated dose settings. Doctors can divide a hand-drawn area into several sub-regions on the human-computer interaction terminal and set target dose values ​​for each. The system automatically generates a variable-dose scanning path, with smooth energy transitions between sub-regions achieved through gradual speed changes. Furthermore, the energy coupling efficiency coefficient η, in addition to being related to the working distance h, can also be extended to correlate with tissue impedance and temperature parameters. The system reserves an impedance spectrum detection module and an infrared thermal imaging interface for real-time correction of the η value, constructing a multi-parameter coupling efficiency model to achieve closed-loop control of biological effects during treatment.

[0023] 3. Multi-axis flexible actuator The multi-axis flexible actuator consists of an XYZ Cartesian coordinate robot or a six-axis collaborative robotic arm, an end effector quick-change gripper, and a plasma therapy head.

[0024] Multi-axis flexible actuators are responsible for performing high-precision automated treatment actions. The plasma treatment head is driven by a multi-axis robot (such as an XYZ Cartesian coordinate system or a six-axis robotic arm) with micron-level positioning accuracy; combined with an end effector floating conformal mechanism and real-time feedback, it can adapt to changes in curvature and minute undulations of the human body surface to achieve smooth scanning.

[0025] The motion characteristics of the multi-axis flexible actuator include, but are not limited to: supporting S-shaped speed curve planning and having micron-level repeatability positioning accuracy; the end-effector quick-change fixture is designed with a floating conforming mechanism, which, together with laser ranging feedback, can adapt to the slight curvature changes of the human body surface.

[0026] 4. Active safety monitoring module The proactive safety monitoring module includes a dual monitoring mechanism: global visual monitoring and local distance closed-loop monitoring. Global visual monitoring: During treatment, the baseline image is compared with the real-time image at a high frame rate (≥15fps) to calculate the displacement vector of feature points. If the displacement exceeds a preset safety threshold (e.g., ±2mm), a pause is triggered immediately.

[0027] Local distance closed loop: Laser ranging data is fed back to the motion controller at a frequency of kHz, and the Z-axis height is adjusted in real time through PID algorithm to maintain a constant working distance (e.g., 10mm ± 1mm).

[0028] The proactive safety monitoring module is also equipped with a tiered safety response mechanism, which implements differentiated handling strategies based on the displacement amplitude and duration: Level 1 response (slight displacement, such as ±1-2mm and duration <0.5s): The system automatically pauses scanning, initiates image relocking and trajectory realignment, and automatically resumes treatment after stabilization, without manual intervention; Level 2 response (moderate displacement, such as ±2-5mm or duration ≥0.5s): triggers audio-visual prompts, and recovery is only possible after confirmation by a doctor; Level 3 response (large displacement, such as >5mm or continuous displacement): immediately execute a hard stop, cut off the plasma output and lock the robotic arm, and generate a displacement event record for postoperative review.

[0029] The active safety monitoring module serves to construct a dual safety protection mechanism to cope with dynamic medical environments. Globally, it monitors patient surface displacement through high-frequency visual comparison to prevent large-scale misalignment; locally, it uses laser ranging closed-loop to adjust the Z-axis height in real time to maintain a constant working distance. Once abnormal displacement or distance deviation is detected, such as displacement trigger threshold ±2mm, distance deviation ±1mm, or exceeding the upper / lower speed limit, an emergency stop is triggered within milliseconds to ensure treatment safety.

[0030] (II) Method Implementation Steps The present invention also provides a control method based on the above system, specifically including the following steps: Step S1: Unstructured intent tagging Doctors use a medical marker to draw a closed outline of any shape around the lesion on the patient's skin. This outline does not need to be a regular geometric shape, but only needs to clearly define the treatment boundary.

[0031] Step S2: Visual capture and semantic parsing The system starts the multimodal visual perception subsystem to acquire the original RGB image.

[0032] Preprocessing: Convert the image to the HSV color space to remove the effects of uneven lighting.

[0033] Feature extraction: Set a dynamic threshold range to extract the marker color and generate a binary image; use morphological dilation and erosion operations to connect broken lines, fill internal voids, and form a complete binary treatment area mask.

[0034] Contour fitting: Extract the set of coordinate points of the outer boundary of the mask, and use a polygon fitting algorithm to simplify the amount of data while retaining key feature points.

[0035] Step S3: Spatial mapping and parameter decoupling Coordinate transformation: Call the hand-eye calibration matrix Map the pixel coordinate set (u,v) to the robotic arm's Cartesian coordinate set (X,Y,Z). base ).

[0036] Path generation: Based on the mapped boundaries, a full-coverage scan path is generated using either the Boustrophedon decomposition algorithm or the spiral algorithm.

[0037] Speed ​​planning: Based on the preset treatment dose Plasma output power P and effective plume diameter Using the formula Calculate the target velocity vi of each discrete point i on the path and generate a "position-velocity" synchronized trajectory table.

[0038] For treatment areas with dramatic curvature changes (such as the shoulder and joints), the system supports a 3D surface path planning mode. This mode uses binocular vision or structured light sensors to reconstruct a 3D point cloud model of the treatment area, projects a 2D treatment area mask onto the 3D surface, and generates an equidistant offset path, ensuring that the treatment head axis is always aligned with the local normal direction of the body surface. For periodic physiological movements (such as breathing), the system incorporates motion prediction and synchronous tracking algorithms. By analyzing historical displacement trajectories, it predicts the movement trend of the next cycle, achieving "soft following" control and avoiding frequent abrupt stops triggered by minute displacements.

[0039] Step S4: Human-machine collaboration confirmation The identified treatment area outline and the generated planned path are overlaid on the interactive terminal screen. Doctors can manually fine-tune the boundaries or click the "Start" command after confirming that everything is correct.

[0040] Step S5: Adaptive Execution and Dynamic Monitoring Motion execution: The robotic arm moves the treatment head according to the trajectory, and the motion controller synchronously controls the start and stop of the plasma generator (gas is emitted when it moves and turned off when it stops).

[0041] Constant distance control: The laser sensor provides real-time feedback on the distance error Δh, and the Z-axis servo motor compensates in real time to ensure that the distance between the output port and the body surface remains constant.

[0042] Displacement interruption: The vision system runs a displacement detection algorithm in parallel, calculating the cross-correlation coefficient or Euclidean distance between the current frame and the reference frame. If an unexpected displacement of the patient's body surface is detected (e.g., respiratory fluctuations exceeding the compensation range or body position movement), the system triggers an emergency stop signal within milliseconds, cutting off the high-voltage output and locking the robotic arm, resuming operation after recalibration.

[0043] Furthermore, this system also includes a dose-motion dynamic coupling control algorithm. Traditional equipment typically employs a constant scanning speed, which cannot adapt to dose unevenness issues caused by power fluctuations or complex curved surfaces. This algorithm constructs an energy deposition physical model, using the "target biological dose" as an input variable for the control system, and solves in real time the "optimal scanning speed" at each path point, achieving precise closed-loop control of the dose. It also establishes a real-time dynamic mapping model between plasma therapy dose and robotic arm scanning speed.

[0044] The implementation path of the dose-motion dynamic coupling control algorithm includes: 1. Physical Model Construction Based on the energy transfer mechanism of the interaction between plasma jet and biological tissue, the effective therapeutic dose D per unit area (Unit: J / cm²) is... 2 The plasma effective power (or equivalent biological effect unit) is defined as the time integral of the plasma effective power over a unit area.

[0045] In continuous scanning mode, assuming the effective area of ​​the plasma plume on the body surface is rectangular or elliptical, its equivalent width along the direction of motion is... (Unit: cm), with the length perpendicular to the direction of motion being L. When the treatment head moves at an instantaneous velocity v(t), the time Δt during which a point PP on the body surface is affected by plasma can be expressed as:

[0046] Therefore, the cumulative dose D received at this point is related to the effective plasma output power. The relational model is as follows:

[0047] in: The effective electrical power output of the plasma generator (unit: W) is obtained in real time from feedback by the high-frequency power supply. η: Energy coupling efficiency coefficient (0<η≤1), which is related to the distance h from the output port to the body surface, the composition of the ambient gas and the tissue impedance. The functional relationship is obtained through experimental calibration. Effective spot area in a single action ; v: Instantaneous tangential scanning velocity of the treatment head relative to the body surface (unit: cm / s).

[0048] After simplification, the steady-state dose-velocity coupling equation used in this system is obtained:

[0049] in, These are the system's geometric constants.

[0050] 2. Speed ​​Inverse Solution and Dynamic Programming Logic To achieve a constant target dose The control system needs to be based on real-time monitoring. And h, dynamically calculate the target velocity at each moment. The specific steps include: (1) Real-time parameter acquisition Power feedback: Read the current actual output power from the plasma power controller. (Sampling frequency ≥ 1kHz) to eliminate the influence of power supply fluctuations.

[0051] Distance compensation: real-time working distance is read from the end laser rangefinder. Look up the table to obtain the corresponding coupling efficiency correction coefficient. .like Exceeding the safety threshold If this happens, an emergency stop will be triggered, and speed calculation will not be performed.

[0052] (2) Target velocity calculation Substituting the above real-time parameters into the inverse function of the coupling equation, the theoretical target velocity is calculated:

[0053]

[0054] (3) Kinematically constrained smoothing Because the robotic arm has the maximum acceleration And the maximum jerk (jerk) limit, applied directly. This could lead to mechanical vibration. Therefore, an S-shaped velocity curve smoothing filter is introduced: If calculated Then the target speed will be limited to .

[0055] At the same time, set a minimum safe speed. (To prevent localized burns due to excessively slow speed) and maximum efficiency speed Final execution speed for:

[0056] (4) Power-speed coordinated compensation When encountering complex curved surfaces or when the robotic arm reaches its speed limit (e.g.) If the required dose is still insufficient (i.e., the required dose is too high, even at the slowest speed), the algorithm automatically triggers a power derating mechanism:

[0057] Reverse adjustment of plasma power supply setting: .

[0058] Achieving "power adaptation under speed constraints" ensures that the actual deposition dose is consistent under any motion conditions. Always approaching .

[0059] The technical advantages of establishing a dose-rate dynamic coupling control algorithm are:

[0060] Enhanced anti-interference capability: through real-time introduction and Feedback: The algorithm can automatically offset the effects of power supply voltage fluctuations, slight changes in airflow, and distance changes caused by human respiration on the treatment dose, keeping the dose error within ±5%.

[0061] Enhanced adaptability to curved surfaces: When dealing with uneven human body surfaces, the coupling efficiency changes with distance h. When changes occur, the algorithm automatically adjusts the speed v to compensate, without requiring manual intervention from the doctor, thus ensuring the uniformity of the curved surface treatment.

[0062] Enhanced treatment safety: Built-in The limitation and power inverse regulation mechanism fundamentally eliminates the risk of local overheating and burns caused by the robotic arm jamming or excessively slow speed.

[0063] III. Beneficial Effects

[0064] Compared with the prior art, the present invention has the following significant inventive advantages: 1. It pioneered a new paradigm of "clinical intuition-driven machine execution". Breaking away from the limitations of traditional medical equipment that relies on complex software modeling, this innovative approach directly transforms the doctor's most familiar "circle-drawing" action into machine instructions. This instant conversion capability from unstructured input to structured output significantly lowers the clinical usage threshold of automated equipment, enabling true "zero-training" operation.

[0065] 2. Achieved "algorithm-level standardization" of treatment dosage. By dynamically coupling medical dosage parameters with mechanical motion parameters (speed, spacing), the uncertainty in treatment caused by differences in human experience is eliminated. Regardless of the irregularity of the lesion shape, the system can ensure a high degree of consistency in energy deposition per unit area, establishing a new standardized benchmark for plasma therapy.

[0066] 3. A proactive defense system for flexible targets has been constructed: Unlike the "blind running" mode of industrial robots, this invention introduces a dual safety mechanism of "global visual monitoring + local laser closed-loop control." It not only adapts to the curvature of the human body to maintain a constant distance, but also sensitively detects the patient's minute displacements and responds instantly, solving the safety challenges of using automated equipment in dynamic medical environments.

[0067] 4. Possesses scalability and robustness for complex clinical scenarios. Based on its core architecture, this invention integrates extended functional modules such as deep learning-assisted recognition, 3D curved surface path planning, graded safety response, multi-treatment head collaboration, and closed-loop efficacy monitoring, forming a clearly structured, customizable, and upgradeable technical system. The system can adapt to complex backgrounds such as scars, pigmentation, and ointment residue, and is compatible with dynamic environments such as respiratory movements and curved surface deformations. Through a graded safety mechanism, it minimizes treatment interruptions while ensuring absolute safety, significantly improving product feasibility and clinical application scope. Attached Figure Description

[0068] Figure 1 System overall architecture diagram This diagram illustrates the connection between vision, decision-making, and execution. Detailed Implementation

[0069] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. These embodiments are intended to explain the technical principles of the present invention, rather than to limit its scope of protection.

[0070] Example 1: System Hardware Architecture This embodiment details a hardware implementation scheme for a plasma adaptive scanning therapy system based on unstructured visual guidance. The system adopts a modular design, mainly consisting of four parts: a multimodal visual perception subsystem, an intelligent decision-making and planning center, a multi-axis flexible actuator, and an end-effector integrated treatment component. Each module interacts with real-time industrial Ethernet via a high-speed bus to ensure microsecond-level synchronous control accuracy.

[0071] 1. Multimodal Visual Perception Subsystem This module is responsible for acquiring high-fidelity images of the patient's body surface and is the physical basis for the "circle recognition" logic.

[0072] 1.1 Industrial camera: Example model Basler ace 2 acA2440-75um (or equivalent global shutter camera).

[0073] Key specifications: 2448×2048 resolution (5 megapixels), Sony IMX264 (CMOS) sensor, 75fps (at full resolution), GigE interface (gigabit Ethernet).

[0074] Reasons for selection: The global shutter feature can eliminate image distortion (rolling-up curtain effect) during the movement of the robotic arm; the high frame rate supports real-time displacement monitoring; the high resolution ensures that it can recognize marker lines with a width of only 1-2mm.

[0075] 1.2 Optical Lens: Example model: Computar M0814-MP2 (8mm fixed focal length lens).

[0076] Key parameters: 8mm focal length, F1.4 aperture, supports 2 / 3” target surface, low distortion (<0.1%).

[0077] Structural relationship: The lens is mounted on the camera's C-mount and fixed vertically downward to the center of the gantry beam. The optical axis is strictly perpendicular to the plane of the treatment bed, and the field of view (FOV) covers a treatment area of ​​400mm × 300mm.

[0078] 1.3 Spectral Enhancement Light Source: Example model: CCS LDR2-100SW Ring LED Light Source (or Custom Wavelength Light Source).

[0079] Key parameters: The emission color can be selected as blue (470nm) or white, the brightness is adjustable (0-100%), and the flicker controller is triggered synchronously.

[0080] Innovative Design: Specific wavelengths of light are selected for medical markers (typically blue or black) to maximize the contrast between ink and skin. The light source is coaxially mounted around the lens to eliminate shadows and ensure image illumination uniformity >90%.

[0081] 2. Intelligent Decision-Making and Planning Hub This unit is the "brain" of the system, responsible for image processing, path planning, and issuing motion control commands.

[0082] 2.1 Main Controller (IPC): Example models: Advantech IPC-610L industrial PC or NVIDIA Jetson AGXOrin (capable of heavy AI inference).

[0083] Configuration: Intel Core i7-12700 processor, 32GB DDR4 RAM, 512GB NVMe SSD, dedicated RTX A2000 graphics card (for accelerating image segmentation algorithms).

[0084] Functionality: Runs the Ubuntu Linux system and deploys OpenCV, PCL (point cloud library), and self-developed path planning algorithm software.

[0085] 2.2 Motion control card: Model examples: Googol Technology GT-800-PGT-PCIe (Googol Technology multi-axis motion control card) or Beckhoff CX2030 (EtherCAT master).

[0086] Key parameters: Supports 4-8 axis linkage, interpolation cycle ≤1ms, supports S-shaped speed curve planning, and has hardware limit and emergency stop input interfaces.

[0087] Connection relationship: Communicates with the main controller via PCIe or EtherCAT bus, receives the planned trajectory points, and sends pulse / bus commands to the driver.

[0088] 2.3 Human-Computer Interaction Terminal (HMI): Example model: 15-inch industrial-grade capacitive touchscreen (IP65 protection rating).

[0089] Functions: Displays real-time video stream, recognizes contour overlays, provides parameter settings, and includes an emergency stop button.

[0090] 3. Multi-axis Flexible Actuator This embodiment adopts a gantry-type XYZ three-cartesian coordinate robot structure, which has better rigidity, higher repeatability and positioning accuracy and simpler control algorithm compared with a six-axis robotic arm, making it suitable for planar scanning tasks.

[0091] 3.1 X / Y axis module (horizontal scan): Model examples: HIWIN KK100C series linear module or MISUMI EZS series.

[0092] Key parameters: X-axis travel 600mm / Y-axis 400mm, lead 10mm / 20mm, repeatability ±0.02mm, maximum speed 1000mm / s, load capacity ≥5kg.

[0093] Drive motor: Panasonic MINAS A6 series AC servo motor (400W), equipped with a 23-bit absolute encoder.

[0094] Structural Assembly: The Y-axis module is mounted horizontally on both side columns, and the X-axis module is mounted on the Y-axis slide, forming a cantilever gantry structure. The base is made of heavy-duty aluminum profile welded and subjected to aging treatment to ensure stability under high-speed movement.

[0095] 3.2 Z-axis module (vertical lifting): Model example: HIWIN KK50C series small linear module.

[0096] Key parameters: 200mm stroke, repeatability ±0.01mm, with self-locking function (to prevent slippage during power failure).

[0097] Drive motor: Panasonic servo motor (200W).

[0098] Structural assembly: Vertically mounted on the front end of the X-axis slide, used to adjust the height of the treatment head and follow the curves of the human body.

[0099] 3.3 Cable Carrier System: Model example: IGUS E2 series miniature cable chain.

[0100] Function: It accommodates and protects all motor power lines, encoder lines, air hoses and sensor signal lines, ensuring that they do not become tangled or worn during high-speed reciprocating motion.

[0101] 4. End-Effector Integrated Therapy Assembly This component is a core part that directly affects the patient, integrating treatment and sensing functions.

[0102] 4.1 Quick-Change Fixture Structure: It adopts a pneumatic gripper or manual spiral locking structure, which is compatible with the outer diameter of mainstream plasma therapy handpieces (e.g., Φ30mm).

[0103] Material: Medical grade stainless steel (304) or PEEK material, resistant to disinfection and corrosion.

[0104] Degrees of freedom: The clamp is designed with a fine-tuning knob to adjust the pitch and roll angles of the handle, ensuring that the handle axis is always perpendicular to the treatment plane.

[0105] 4.2 Laser rangefinder sensor Model examples: Keyence LJ-V7000 series or Micro-Epsilon optoNCDT 2300.

[0106] Key parameters: The measurement principle is laser triangulation reflection method, the measurement range is 10-50mm, the resolution is 1μm, and the sampling frequency is 2kHz.

[0107] Installation location: Install parallel to the output port (distance <10mm), with the beam projection point coinciding with the plasma action point or having a known offset.

[0108] 4.3 Plasma Therapy Handpiece Interface: The high-voltage power cord and air supply hose of the handle are introduced through a cable chain, and the plasma treatment head is connected to the external plasma generator host.

[0109] Synchronization signal: The handle control line is connected to the DO (digital output) port of the motion control card to achieve hard-wired synchronization of "air / discharge when moving, and shutdown when stopping", with a delay of <10ms.

[0110] 5. Safety and Auxiliary Systems Emergency stop circuit: A dual-channel hard-wired emergency stop circuit conforming to ISO 13850 standard, connected in series with the HMI emergency stop button, robotic arm limit switch and software watchdog signal, to cut off servo enable and plasma high voltage output.

[0111] Audible and visual alarm tower: Three-color lights (green - running, yellow - standby / paused, red - fault / emergency stop) and a buzzer to visually display the system status.

[0112] Gas piping assembly: Includes a precision pressure reducing valve, flow meter and solenoid valve to ensure a stable flow of working gas (including helium, argon or air), and the piping uses medical-grade silicone tubing.

[0113] 6. Summary of Structural Layout The overall equipment has a "gate" shaped structure. The base is fixed to the ground or a support next to the treatment bed; the gantry beam spans above the treatment area; the visual camera is centrally suspended; and the XYZ module drives the end treatment head to move freely in the space below the beam. This layout ensures a sufficient treatment range (600×400mm), avoids obstructing the doctor's operation with the main body of the equipment, and facilitates the rapid removal of patients in emergencies.

[0114] 7. Multi-treatment head collaboration and monitoring expansion interface To improve the treatment efficiency of large-area lesions, the system reserves a multi-treatment head collaborative control interface, supporting simultaneous scanning of 2-4 plasma treatment heads. The effective plume width *w* in the dose-velocity coupling model is extended to the sum of the effective plume widths of each treatment head multiplied by a coordination coefficient. The path planning module automatically generates a multi-machine collaborative trajectory, avoiding interference and overlap. The output port reserves installation interfaces and data channels for hyperspectral imaging, fluorescence imaging, or optical coherence tomography (OCT) modules, supporting the simultaneous acquisition of tissue response parameters during treatment, achieving an upgrade from "open-loop dose control" to "closed-loop efficacy control."

[0115] Through the careful selection and systematic integration of the aforementioned components, this embodiment constructs a high-precision, high-response, and high-safety hardware platform, providing a solid physical carrier for the aforementioned software algorithms such as "visual guidance," "dose coupling," and "active safety," and fully meeting the requirements for the disclosure of technical solutions.

[0116] Example 2: Detailed Explanation of the Core Algorithm Flow This embodiment details the core software algorithm flow running in the Intelligent Decision Center (IPC). This flow is crucial for realizing the three innovative functions of "unstructured marker recognition," "dose-motion dynamic coupling," and "active safety monitoring." The algorithm is developed using C++, calls the OpenCV computer vision library and a self-developed motion planning module, and runs in a Linux real-time kernel environment, ensuring a control cycle of less than 1ms.

[0117] 1. Intelligent recognition and extraction algorithm for unstructured tags This step aims to transform the irregular, discontinuous lines hand-drawn by doctors into closed area data that machines can understand.

[0118] Step S1.1: Image preprocessing and color space conversion

[0119] Input: Raw RGB images captured by an industrial camera .

[0120] Processing: Since the colors of medical markers (usually blue, black, and purple) are easily affected by skin color and bloodstains in the RGB space, the algorithm first converts the image to the HSV color space (H, S, V).

[0121] Threshold segmentation: Based on a preset ink color model, a dynamic threshold range is set. , , ; For example, regarding blue ink: , , .

[0122] Generate a binarized mask image The marked area has a pixel value of 255, while the background has a value of 0.

[0123] Adaptive Enhancement: The V channel is enhanced using the CLAHE (Contrast Limiting Adaptive Histogram Equalization) algorithm to solve the problem of local recognition failure caused by uneven lighting in the treatment room.

[0124] Step S1.2: Morphological Repair and Contour Closure Problem: The lines drawn by the doctor may have breaks or uneven thickness.

[0125] deal with: (1) Opening: Use a 3x3 structuring element to perform erosion followed by dilation to remove noise (e.g., skin spots).

[0126] (2) Closing: Use 5x5 or larger structuring elements to expand and erode, connecting broken lines and filling tiny gaps.

[0127] (3) Hole Filling: The FloodFill algorithm is used to fill the holes inside the closed contour to form a solid treatment area mask. .

[0128] Step S1.3: Contour extraction and feature selection Algorithm: Extracting topology using the Suzuki-Abe topology analysis algorithm (findcontours) All external contours.

[0129] Filtering logic: Calculate the area and perimeter of each contour.

[0130] Remove areas that are too small ( (This is considered noise) or an excessively large (outside the field of view) outline.

[0131] Calculate roundness Optional, to help determine whether it is a valid tag.

[0132] Output: The set of coordinates of the largest and most reasonable closed contour. .

[0133] 2. Hand-eye calibration and 3D coordinate mapping algorithm This step converts two-dimensional pixel coordinates into three-dimensional physical coordinates for the robotic arm, serving as a bridge connecting vision and execution.

[0134] Step S2.1: Solving the hand-eye calibration matrix Calibration method: Use the classic nine-point calibration method. The calibration accuracy should not exceed 0.5mm; for example, the reprojection error should be <0.1mm.

[0135] The robotic arm's end effector, carrying a calibration probe, sequentially touches nine known physical coordinate points on a calibration plate. ; The camera captures an image at the corresponding location, and the pixel coordinates of the needle tip are extracted. ; Solving affine transformation matrices using the least multiplication method ; .

[0136] Step S2.2: Real-time coordinate transformation Execution: The set of contour points output in step S1.3... Substitute into matrix The physical contour point set in the coordinate system of the robotic arm base is obtained. ; Boundary simplification: Since the number of points N in the original contour is huge, the Douglas-Peucker algorithm is used to fit the polygon. While maintaining the shape accuracy (error <0.5mm), the contour is simplified into a polygon containing 50-100 key vertices, reducing the computational load of subsequent path planning.

[0137] 3. Dose-driven dynamic path planning algorithm This is the core innovation of the invention, realizing the transformation from "geometric path" to "treatment process".

[0138] Step S3.1: Parameter Input and Model Building Input parameters: Target therapeutic dose (J / cm) 2 (This is set by the doctor in the HMI.)

[0139] : Plasma generator output power (W), read from device status.

[0140] The effective treatment diameter (cm) of the plasma plume was obtained through preliminary experimental calibration (e.g., 0.5cm).

[0141] Energy coupling efficiency coefficient (empirical value, usually 0.8-0.9).

[0142] Speed ​​calculation model: According to the principle of conservation of energy, the energy received per unit area For continuous scanning, , .

[0143] Derive the scan speed formula:

[0144] Step S3.2: Boustrophedon Decomposition (Full Coverage Path Generation) Domain decomposition: This involves decomposing the simplified polygonal region. Perform trapezoidal decomposition or directly adopt a zig-zag filling strategy.

[0145] Path spacing calculation: To ensure no omissions and uniform energy distribution, the path row spacing is set. The overlap coefficient α is usually taken as 0.7-0.8 (i.e. 20%-30% overlap).

[0146] Trajectory generation: Determine the main scanning direction (usually along the long axis of the polygon to reduce idle travel).

[0147] Generate a series of parallel line segments, with the endpoints of the line segments obtained by trimming the polygon boundaries.

[0148] Add arc blending or S-shaped acceleration / deceleration sections at both ends of the line segment to avoid sudden speed changes and dose accumulation caused by right-angle turns.

[0149] Output: Generate a time-series trajectory point set containing position (x, y), velocity v, and acceleration a. .

[0150] 4. Active safety monitoring and closed-loop control algorithm This step is performed in parallel during the treatment process to ensure safety in a dynamic environment.

[0151] Step S4.1: Visual Global Displacement Monitoring (Visual Servoing for Safety) Benchmark establishment: Before treatment begins, store the benchmark image feature point set. (e.g., ORB or SIFT features).

[0152] Real-time tracking: Images are acquired at a frequency of 10-15 fps during treatment, and the current feature point set is extracted. .

[0153] Pose calculation: RANSAC algorithm for matching. and Calculate the homography matrix H and solve for the translation amount. , and rotation amount .

[0154] Judgment Logic: If (e.g., 2mm) or (e.g., 5º); Action: Immediately send an Emergency Stop signal to the motion control card, shut down the plasma output, and display an alarm "Patient displacement too large" in the HMI pop-up window.

[0155] Step S4.2: Laser Local Constant Distance Closed-Loop Control (PID Distance Control) Sampling: The laser rangefinder reads the current distance at a frequency of 2kHz. .

[0156] Error calculation: ,in Set the preset working distance (e.g., 10mm).

[0157] PID control:

[0158] Where u(t) is the speed correction amount of the z-axis motor. Execution: Superimpose u(t) onto the currently planned z-axis command to drive the z-axis to float up and down in real time, compensating for human breathing fluctuations or changes in surface height, and ensuring e(t) < 0.2mm.

[0159] Step S4.3: Software and hardware synchronization triggering Mechanism: When the motion control card reaches each trajectory point (xk, yk), it sends a TTL pulse signal to the plasma host through the hardware GPIO port.

[0160] The control logic includes: Robotic arm speed > 0 → Output High level (plasma is turned on).

[0161] Robotic arm speed = 0 (or emergency stop) → Output low level (plasma off).

[0162] Delay control: Signal delay is guaranteed to be <1ms through FPGA or real-time kernel to prevent uneven dosage during start-up and shutdown.

[0163] 5. Summary of Algorithm Flowcharts Initialization: Load calibration parameters and start the camera and sensor.

[0164] Human-machine confirmation: The path is displayed, and the doctor is waiting to click "Start".

[0165] Real-time Loop: Send XY axis motion command + plasma switch signal.

[0166] End: Reset the robotic arm and generate a treatment report.

[0167] Through the detailed algorithm description above, this embodiment fully discloses how to transform a doctor's intuitive intentions into precise, safe, and standardized machine actions, demonstrating the high level of creativity and technical feasibility of this invention at the software logic level.

[0168] Example 3: Typical Clinical Procedure This embodiment aims to illustrate in detail the complete operation steps from patient preparation to treatment completion in a specific clinical scenario, so as to intuitively demonstrate how the system of the present invention can operate efficiently and safely in a real medical environment.

[0169] Step 1: Pre-treatment preparation and system initialization 1. Patient positioning: Guide the patient to lie flat on the treatment bed, adjust the position according to the location of the lesion, and ensure that the patient is comfortable and can remain still. If necessary, use soft pillows or restraints for additional fixation.

[0170] 2. Lesion marking: After routine disinfection, the doctor uses a sterile medical marker (blue or black is recommended for high contrast with skin tone) to draw a closed, continuous outline around the lesion on the patient's skin surface freehand. This outline should clearly define the treatment boundary and can be drawn based on anatomical landmarks or the doctor's clinical judgment; a regular geometric shape is not required.

[0171] 3. System Startup and Zeroing: The operator starts the system power and executes the "System Initialization" command through the human-machine interface (HMI). The motion control system drives the XYZ platform to automatically find the mechanical origin, complete the zeroing operation, and establish the absolute coordinate system.

[0172] 4. Parameter settings: The doctor enters treatment parameters on the HMI, which mainly include: Target therapeutic dose D set (Unit: J / cm) 2 (This is based on the pathological type and treatment plan.)

[0173] Working distance h set (Unit: mm), usually the optimal working distance of the output port, such as 10 mm.

[0174] Confirm that the pressure and flow rate of the plasma gas source (e.g., helium) are normal.

[0175] Step 2: Visual Recognition and Path Planning (Automated Execution) 1. One-click capture: The doctor clicks the "Capture Image" button on the HMI. The multimodal visual perception subsystem is immediately triggered, and a high-resolution industrial camera captures a clear image of the treatment area, including hand-drawn markings, under spectrally enhanced illumination.

[0176] 2. Intelligent analysis and display: The intelligent decision-making center completes image processing in milliseconds.

[0177] It automatically identifies and extracts hand-drawn outlines, repairs minor breaks, and generates a closed treatment area mask.

[0178] The coordinates of the two-dimensional image are converted into the three-dimensional physical space coordinates of the robotic arm through the hand-eye calibration matrix.

[0179] According to the set target dose D set Given the current plasma power P, the system automatically calculates the zigzag scanning path required to cover the entire region and assigns the optimal scanning velocity v to each point on the path, generating a position-velocity synchronization trajectory table.

[0180] 3. Path Preview and Confirmation: The system overlays the identified treatment area outline (e.g., a red highlighted line) and the generated scan path (e.g., a green dashed line) onto the real-time video feed of the HMI. Doctors can visually verify whether the system accurately understood their marking intent. For example, if correct, click "Confirm Path"; if there are minor deviations in the boundaries, the outline can be easily corrected by dragging the outline with a finger on the touchscreen.

[0181] Step 3: Adaptive Treatment Execution (Automated Execution) 1. Initiating Treatment: After confirming that all personnel are in a safe area, the doctor clicks the "Start Treatment" button on the HMI. The system then enters the fully automated execution phase.

[0182] 2. Dynamic scanning and dose control: The robotic arm, following the planned "position-velocity" trajectory, drives the plasma therapy head to begin moving.

[0183] Synchronous control: The motion control card controls the plasma generator in real time through the hardware I / O port to ensure that the plasma plume is ejected immediately when the robotic arm moves; the plume is shut off immediately when the robotic arm pauses or completes its trajectory to prevent ineffective irradiation.

[0184] Constant distance control: The laser rangefinder at the end of the device monitors the distance h between the output port and the skin in real time at a frequency in the kHz range. Once a minute change (Δ) caused by breathing or surface curvature is detected... h The Z-axis servo motor responds immediately, and the height is finely adjusted in real time through a PID algorithm to ensure that the working distance remains constant at h. set Within ±0.2mm range.

[0185] Power-velocity coupling: The system reads the actual output power P of the plasma power source in real time. If the power fluctuates slightly, the control system will adjust the output power according to the core formula v=k·P / D. setDynamically fine-tuning the scan rate of subsequent path points and automatically compensating for power variations ensures that the actual deposition dose always closely approximates D. set .

[0186] Step 4: Proactive security monitoring (parallel automated execution) 1. Dual Safety Protection: Throughout the treatment process, two safety mechanisms operate in parallel, with higher priority than motor control commands: Global visual monitoring: The vision system continuously acquires images at a frame rate of ≥15fps. The algorithm compares the current image with the baseline image before treatment begins in real time and calculates the displacement vectors of feature points.

[0187] Laser local monitoring: The distance closed-loop control program continuously monitors the distance error e(t).

[0188] 2. Abnormal Response and Emergency Stop: Scenario A (Patient Micro-movement): If the patient experiences slight trunk displacement (e.g., <2mm) due to coughing, and the visual global monitoring system detects the displacement but it is still within the compensable range, the system will pause scanning and issue a voice prompt "Please remain still," and will automatically resume scanning after the image stabilizes again.

[0189] Scenario B (Emergency Avoidance): If the patient's body displacement exceeds the preset safety threshold (e.g., >2mm) or the laser rangefinder shows a distance exceeding the safety range (e.g., >12mm), the system will immediately trigger a hard-wired emergency stop: within <10ms, the plasma high-voltage output will be cut off, all servo motors will be locked, and a red alarm window will pop up on the HMI saying "Patient displacement is too large, treatment interrupted!" At the same time, the audible and visual alarm tower will sound an alarm.

[0190] Scenario C (Operator Intervention): In any situation, if the operator or doctor presses any emergency stop button, the system will immediately cut off all power output with the highest priority.

[0191] Step 5: End of treatment and repositioning 1. Automatic completion: After the robotic arm completes all trajectory points, it automatically returns to the safe zero position and the plasma generator is turned off.

[0192] 2. Report generation: The system automatically generates a report for this treatment, including treatment time, coverage area, estimated total dose, average speed, abnormal event records, etc., for doctors to archive.

[0193] 3. Patient evacuation: After checking the treatment area and confirming that everything is in order, the doctor will assist the patient to leave the treatment area.

[0194] Through the above process, doctors only need to complete two core actions: "drawing a circle" and "confirming with a button." All other complex tasks, such as path planning, precision control, dosage adjustment, and safety assurance, are automatically handled by the system. This fully embodies the "what you see is what you get" design concept and highly intelligent clinical applicability of this invention.

[0195] The above figures and embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention. All such modifications or substitutions should be covered within the scope of the claims of the present invention and do not constitute any limitation on the protection scope of the present invention.

Claims

1. A plasma adaptive scanning therapy system based on unstructured visual guidance, characterized in that, include: The multimodal visual perception subsystem includes a high-resolution global shutter industrial camera, a specific band spectral enhancement light source, and a laser rangefinder, used to acquire images of the treatment area including hand-drawn markings by doctors, and to monitor the micro distance between the treatment head and the body surface in real time. The intelligent decision-making and planning hub includes a built-in image semantic segmentation engine, a hand-eye coordinate mapping module, a dose-motion coupling solver, and a dynamic path planner. It is used to identify unstructured hand-drawn contours, map them into three-dimensional physical coordinates, and dynamically plan the full-coverage scanning path and calculate the optimal motion speed based on the dose-motion coupling model. The multi-axis flexible actuator consists of a multi-axis robot, an end effector quick-change gripper, and a plasma therapy head, and is used to drive the plasma therapy head to perform high-precision automated treatment actions; The active safety monitoring module includes a dual monitoring mechanism of global visual monitoring and local distance closed-loop monitoring, which is used to monitor the patient's body surface displacement and the distance of the treatment head in real time during treatment and trigger a response when abnormalities occur.

2. The plasma adaptive scanning therapy system based on unstructured visual guidance according to claim 1, characterized in that, The multimodal visual perception subsystem uses a specific band spectral enhancement light source with a wavelength complementary to the color of the medical marker to maximize the contrast between the marking and the skin background.

3. The system according to claim 1, characterized in that, The intelligent decision-making and planning center includes a dose-motion coupling solver, which is based on a treatment dose model. , Where D is the treatment dose, P is the power, η is the energy coupling efficiency, v is the scan rate, and w is the effective width of the plume; The solver dynamically inversely solves the theoretical scanning speed v of each path point based on the set target dose, current power, and effective plume width.

4. The plasma adaptive scanning therapy system based on unstructured visual guidance according to claim 3, characterized in that, When the plasma therapy head uses two or more output ports to output synchronously, the effective width w of the plume is the sum of the effective widths of the plumes from each output port.

5. The plasma adaptive scanning therapy system based on unstructured visual guidance according to claim 1, characterized in that, The global visual monitoring mechanism of the active safety monitoring module compares the reference image with the real-time image at a high frame rate during the treatment process. If the displacement of the feature point exceeds the preset safety threshold, a pause is immediately triggered.

6. The plasma adaptive scanning therapy system based on unstructured visual guidance according to claim 1, characterized in that, In the local distance closed-loop mechanism of the active safety monitoring module, laser ranging data is fed back to the motion controller at a frequency of kHz, and the Z-axis height is adjusted in real time through a PID algorithm to maintain a constant working distance.

7. A plasma adaptive scanning therapy method based on unstructured visual guidance, employing the system described in any one of claims 1 to 6, characterized in that, Includes the following steps: Step S1: Unstructured intent marking. The doctor uses a medical marker to draw a closed outline of any shape around the lesion on the patient's body surface. Step S2: Visual capture and semantic parsing. The multimodal visual perception subsystem is activated to acquire images. The treatment area mask is extracted through HSV color space conversion, dynamic threshold segmentation, and morphological operations. Step S3: Spatial mapping and parameter decoupling. The hand-eye calibration matrix is ​​called to map the pixel coordinates to the Cartesian coordinates of the robotic arm. The full-coverage scanning path is generated using the ox-plowing or spiral algorithm, and the target velocity of each path point is calculated according to the dose-motion coupling model. Step S4: Human-machine collaboration confirmation. The outline of the treatment area and the planned path are overlaid on the interactive terminal for confirmation or fine-tuning by the doctor. Step S5: Adaptive execution and dynamic monitoring. The robotic arm drives the treatment head to move according to the trajectory, while maintaining a constant working distance through laser ranging feedback, and realizing adaptive scanning treatment through global visual monitoring of patient displacement.

8. The plasma adaptive scanning therapy method based on unstructured visual guidance according to claim 7, characterized in that, The feature extraction in step S2 specifically includes: converting the image to the HSV color space, setting a dynamic threshold range to extract the marker color, and generating a binarized image; using morphological dilation and erosion operations to connect broken lines, fill internal holes, and form a complete binarized treatment area mask.

9. The plasma adaptive scanning therapy method based on unstructured visual guidance according to claim 7, characterized in that, The speed planning in the step S3 specifically includes: calculating a target speed v of each discrete point i on the path according to a preset treatment dose, a plasma output power P and a plume effective diameter i , generating a "position-speed" synchronous trajectory table.

10. The plasma adaptive scanning therapy method based on unstructured visual guidance according to claim 7, characterized in that, The adaptive execution in step S5 includes: the motion controller synchronously controls the start and stop of the plasma generator to achieve gas output when moving and shutdown when stopping; at the same time, the laser sensor provides real-time feedback of the distance error Δh, which is compensated in real time by the Z-axis servo motor to ensure that the distance between the output port and the body surface remains constant.

Images