A CT-CTP registration-guided magnetically controlled microrobot directional thrombolysis system

Abstract

The main objective of this invention is to provide a CT-CTP registration-guided magnetically controlled microrobot-based targeted thrombolysis system, comprising: a CT / CTP image acquisition module for acquiring tomographic anatomical images and perfusion function images of the patient's brain; an image registration and cerebral vascular modeling module for spatial registration and fusion of the tomographic anatomical images and perfusion function images, and for reconstructing a three-dimensional structural model of the cerebral blood vessels based on the acquired cerebral vascular image data; a path planning module for planning the microrobot's path from the vascular inlet to the target thrombus location using the cerebral vascular three-dimensional model and perfusion information; a microrobot magnetic navigation control module for driving the microrobot along the path using a controllable magnetic field generated externally; and a closed-loop feedback control module for real-time monitoring of the microrobot's position and automatic adjustment of magnetic field parameters based on deviations to guide the microrobot to precisely navigate to the thrombus site for targeted thrombolysis. This addresses the problems of inaccurate positioning and high invasiveness in traditional thrombolysis treatment methods.

Description

Technical Field

[0001] This invention relates to the field of medical devices, and in particular to a CT-CTP registration-guided magnetically controlled microrobot directional thrombolysis system. Background Technology

[0002] Ischemic stroke is often caused by cerebral vascular thrombosis, and traditional thrombectomy and thrombolysis methods have many shortcomings. Clinically, endovascular mechanical thrombectomy or intravenous thrombolytic drugs are mainly used, but both methods have limitations: interventional catheters and other instruments are limited by size and rigidity, making it difficult to penetrate deep into the tortuous and small blood vessels in the brain, and improper operation may damage the vessels; while systemic thrombolytic drugs need to be used within a strict time window, and high doses can easily cause serious side effects such as intracranial hemorrhage. Current image-guided methods mainly rely on digital subtraction angiography (DSA) to obtain two-dimensional projection images of the blood vessel lumen, or preoperative imaging such as CT and MRI to determine the approximate location of the obstruction. However, the accuracy of single-modal image guidance is limited: DSA involves radiation exposure and lacks reflection of brain tissue perfusion status; preoperative images and intraoperative anatomy often differ; and the lack of multimodal information fusion leads to insufficient accuracy in determining the extent of ischemic areas and the location of thrombi. In addition, magnetically controlled microrobots, which have emerged in recent years, show promise for minimally invasive thrombectomy, but the navigation methods of current microrobots are still immature. Without high-precision imaging feedback and intelligent control, the precise localization and manipulation of microrobots in complex cerebral vascular networks present significant challenges. The lack of effective path planning algorithms and closed-loop control makes it difficult for robots to autonomously avoid risky areas and accurately reach the target, limiting their application in actual blood vessels. In summary, traditional thrombolysis treatments suffer from inaccurate localization, high invasiveness, or high risks, necessitating an innovative solution integrating advanced image guidance and magnetically controlled robotic technology to address these technical difficulties. This invention addresses the shortcomings of existing technologies by proposing a CT-CTP registration-guided magnetically controlled microrobot-based directional thrombolysis system to achieve precise, efficient, and safe localization and removal of cerebral thrombi. Summary of the Invention

[0003] The main objective of this invention is to provide a CT-CTP registration-guided magnetically controlled microrobot-based targeted thrombolysis system to address the problems of inaccurate positioning, high invasiveness, or high risk associated with traditional thrombolysis treatments. It includes: a CT / CTP image acquisition module, an image registration and cerebral vascular modeling module, a path planning module, a microrobot magnetically controlled navigation control module, and a closed-loop feedback control module. The key features are: the CT / CTP image acquisition module acquires tomographic anatomical images and perfusion function images of the patient's brain; the image registration and cerebral vascular modeling module spatially registers and fuses the tomographic anatomical images and perfusion function images, and reconstructs a three-dimensional structural model of the cerebral blood vessels based on the acquired cerebral vascular image data; the path planning module uses the three-dimensional model of the cerebral blood vessels and perfusion information to plan the microrobot's path from the vessel inlet to the target thrombus location; the microrobot magnetically controlled navigation control module drives the microrobot along the path using a controllable magnetic field generated externally; and the closed-loop feedback control module monitors the microrobot's position in real time and automatically adjusts the magnetic field parameters based on deviations to guide the microrobot to accurately navigate to the thrombus site for targeted thrombolysis.

[0004] Furthermore, the CT / CTP image acquisition module uses spiral CT scanning to acquire plain CT images and CT perfusion (CTP) images of the patient's head. First, non-enhanced CT structural images of the brain are acquired. Then, a contrast agent is injected via a peripheral vein, and a dynamic CT perfusion scan of the target brain region is performed for approximately 50 seconds to obtain a series of time-series CT images and generate cerebral blood flow perfusion parameter maps, including cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT). These parameters are used to identify the ischemic penumbra and infarct core region of brain tissue, thereby providing basic data for determining the location of thrombus occlusion.

[0005] Furthermore, after completing CT perfusion imaging, the image acquisition module performs cerebral angiography to obtain three-dimensional image data of cerebral blood vessels. The cerebral angiography images are obtained by using CT angiography (CTA) scans to obtain three-dimensional CTA image data of the cerebral artery vascular tree, or by using magnetic resonance angiography (MRA) sequences to acquire cranial vascular image data, which are then used by the image registration and cerebral vascular modeling module to reconstruct the anatomical structure model of cerebral blood vessels.

[0006] Furthermore, the image registration and cerebrovascular modeling module uses an elastic registration algorithm to perform high-precision registration and fusion of CT images and CTP perfusion parameter images. The elastic registration algorithm performs spatial alignment of CT images and CTP images through a multi-resolution free deformation model, preferably using the B-spline free deformation algorithm and mutual information as the registration similarity measure to correct the pose differences between different modal images, so that the CT anatomical structures and CTP functional images coincide in a unified coordinate system.

[0007] Furthermore, the image registration and cerebral vascular modeling module segments and reconstructs the acquired CTA or MRA cerebral vascular images in three dimensions to obtain a three-dimensional structural model of the cerebral vascular tree. The module uses a deep learning algorithm to automatically segment the cerebral vascular images, extract the three-dimensional binary model of the blood vessels, and perform morphological processing and centerline extraction on the model to generate a continuous and connected cerebral vascular topology model, which clearly presents the bifurcation structure of intracranial blood vessels and the geometric attributes of each blood vessel segment. The module also combines CTP perfusion parameter information to assign corresponding perfusion state data to the vascular topology model.

[0008] Furthermore, the path planning module abstracts the cerebral vascular topology model into a directed graph structure. Using a heuristic search algorithm A, it plans the optimal path for the microrobot from the entry node to the target thrombus-containing vascular node on the directed graph. During path planning, an evaluation function f(n) = g(n) + h(n) is set, where g(n) is the cumulative path cost from the starting point to the current node n, and h(n) is the heuristically estimated cost from node n to the target. By designing a weighted cost function for the edges of the vascular directed graph, the path cost is calculated by combining factors such as vascular segment length and diameter, and the degree of perfusion abnormality in the blood supply area. Preferably, the path cost is proportional to the vascular length and inversely proportional to the vascular diameter. A penalty weight is added for vascular segments with severe perfusion abnormalities, thereby guiding the algorithm to prioritize vascular paths with larger diameters and normal blood supply. Algorithm A iteratively expands the node with the smallest evaluation function f(n) until the optimal path to the target is found.

[0009] Furthermore, the microrobot magnetic navigation control module includes a micro thrombolytic robot and an external magnetic field generator. The micro thrombolytic robot is sub-millimeter in size, slender cylindrical or spherical, and has directional magnetic elements sealed inside to generate a magnetic moment in a fixed direction. It is coated with biocompatible materials on its surface, enabling it to safely pass through narrow blood vessel segments without damaging the inner wall of the blood vessel. The external magnetic field generator consists of multiple sets of electromagnetic coils distributed around the patient's head, used to generate a three-dimensional adjustable magnetic field and magnetic field gradient to wirelessly drive the movement of the microrobot and control its orientation. Changes in the direction of the magnetic field are used to twist the internal magnetic moment of the robot to align the robot's orientation with the predetermined path direction, and the magnetic force generated by the magnetic field gradient is used to pull the robot forward in the direction of increasing magnetic field strength.

[0010] Furthermore, the external magnetic field generator coordinates and controls the magnetic field vector direction and gradient intensity to achieve attitude adjustment and propulsion control of the microrobot in the vascular network. The magnetic field direction is always oriented towards the direction of the microrobot's current position and the next target path point to guide the robot's magnetic moment to align with the desired course. The magnetic field gradient intensity is adjusted in real time according to the navigation phase: the gradient is increased to provide greater traction when the robot needs to accelerate forward, and the gradient intensity is reduced to decelerate and accurately position when approaching the target thrombus area. This enables the microrobot to smoothly turn along the vascular path, move forward in the center, and accurately stop at the target thrombus for directional thrombolysis.

[0011] Furthermore, the closed-loop feedback control module includes a real-time positioning unit for the microrobot, which combines magnetic signal sensing and medical imaging to monitor the position of the microrobot. The real-time positioning unit acquires the weak magnetic field signal generated by the built-in magnetic moment of the microrobot in real time through a multi-channel magnetic field sensor array arranged around the patient's head, and calculates the spatial coordinate position of the robot. At the same time, it uses medical imaging equipment to verify the position of the robot in the blood vessel. This can be done by using digital subtraction angiography (DSA) to obtain the fluoroscopic position image of the robot, or by using a fast MRI sequence to perform positioning scanning of the robot. The magnetic sensing positioning results and the imaging positioning results are fused and corrected through a data fusion algorithm. Kalman filtering is preferably used to achieve sensor data fusion to improve the accuracy and robustness of the microrobot positioning and provide a high-precision position feedback signal for closed-loop feedback control.

[0012] Furthermore, the closed-loop feedback control module includes a trajectory tracking control unit, which automatically adjusts the magnetic field control input based on the current position deviation of the microrobot to enable the robot to travel along a predetermined path. The trajectory tracking control unit acquires the desired trajectory of the planned path and the current position of the robot, calculates the robot's deviation vector, including position deviation and orientation deviation, and inputs this deviation into the control algorithm to generate magnetic field correction commands. The control algorithm uses a pre-designed performance index function to balance trajectory tracking error and control energy consumption, and uses either a linear quadratic regulator (LQR) control strategy or a model predictive control (MPC) strategy for feedback control of the robot. When using LQR control, the microrobot's magnetic navigation system is linearized and the optimal state feedback gain matrix is ​​solved based on the performance index. The magnetic coil current adjustment is calculated in real time to correct the robot's position deviation. When using MPC control, the magnetic field control sequence is optimized in a rolling manner within a finite prediction time domain based on the robot's kinematic model, and the constraints of magnetic field strength and rate of change are explicitly considered, thereby improving the accuracy, stability, and robustness of trajectory tracking and ensuring that the microrobot safely and efficiently reaches the target location and performs thrombolysis in a complex cerebrovascular environment.

[0013] Compared with the prior art, the present invention has significant beneficial effects:

[0014] Improved image-guided accuracy: By performing high-precision registration and fusion of cranial CT anatomical images and CTP perfusion function images, a comprehensive image containing anatomical structure and blood flow information is obtained, which can accurately locate the ischemic penumbra and the vascular branches where the thrombus is located, ensuring that the robot navigation target is clear and reducing positioning errors.

[0015] Reasonable vascular pathway planning: Based on the three-dimensional topological model of cerebral blood vessels and real-time perfusion parameters, this invention adopts a CT-CTP registration-guided magnetically controlled microrobot directional thrombolysis system algorithm combined with a customized cost function for path planning. It automatically avoids stenotic vascular segments and severely ischemic areas, and the planned path is safer, smoother, and more feasible, reducing the risk of the robot getting lost or blocked in the blood vessels.

[0016] Precise magnetic navigation control: The proposed magnetic navigation strategy utilizes distributed electromagnetic coils to generate an adjustable three-dimensional magnetic field, enabling synchronous remote control of the microrobot's posture and motion. The magnetic field direction is aligned with the robot's next target point in real time, and the dynamic adjustment of the magnetic field gradient provides traction, allowing the robot to move smoothly and turn along the blood vessel centerline with high heading accuracy, enabling it to accurately reach the target thrombus site.

[0017] Robust Closed-Loop Control: This invention constructs a positioning system that integrates magnetic sensing and image fusion, combined with advanced LQR or MPC control algorithms to form a closed-loop feedback control. The controller adjusts the magnetic field input in real time based on robot trajectory deviations, automatically correcting deviations even when the robot deviates due to blood flow interference. Therefore, the system exhibits excellent robustness and adaptability, ensuring the robot strictly follows the planned path and significantly improving navigation stability.

[0018] The system boasts high safety: the microrobot's small size and biocompatible surface pose no risk of mechanical damage to the blood vessel wall; the magnetic field-driven, non-contact method reduces the risk of irritation and puncture to the blood vessel compared to catheter advancement. Closed-loop control's deceleration and obstacle avoidance mechanisms effectively prevent the robot from colliding with the blood vessel wall or entering untargeted branches. Combined with precise positioning and planning, this significantly improves the safety of the surgical procedure.

[0019] Precise Thrombolysis Localization: Guided by detailed brain perfusion images, doctors can precisely navigate a microrobot to the location of the thrombus for targeted thrombolysis. Compared to traditional systemic medication or extensive perfusion, this invention concentrates thrombolytic drugs or physical thrombectomy at the lesion site, resulting in higher thrombolysis efficiency and reduced impact on surrounding normal brain tissue, thus lowering the risk of complications. The entire thrombolysis process is completed under image monitoring, ensuring real-time control and guaranteeing the effectiveness and precision of the treatment.

[0020] In summary, this invention utilizes multimodal medical image fusion and intelligent magnetically controlled robot technology to achieve precise localization, intelligent path planning, accurate navigation control, and safe and efficient thrombolysis of cerebral thrombosis. It overcomes the shortcomings of existing technologies, improves treatment efficacy while reducing surgical risks, and has significant clinical application value. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0022] Figure 2 This is a flowchart illustrating the CT / CTP image acquisition module of the present invention. Detailed Implementation

[0023] This invention provides a CT-CTP registration-guided magnetically controlled microrobot directional thrombolysis system, which comprises: a CT / CTP image acquisition module, an image registration and cerebral vascular modeling module, a path planning module, a microrobot magnetically controlled navigation control module, and a closed-loop feedback control module. The specific implementation of this invention will be described in detail below, combining the functions and processes of each module.

[0024] CT / CTP Image Acquisition Module Design

[0025] The CT / CTP image acquisition module is used to acquire cross-sectional anatomical images and perfusion function images of the brain, providing basic data for subsequent registration and path planning. Preferably, a 64-slice or 128-slice spiral CT scanner is used to acquire plain CT images (non-enhanced) and CT perfusion (CTP) imaging data of the patient's head. First, a conventional CT scan is performed, acquiring cranial images with a 512×512 matrix, approximately 0.5mm pixel spacing, and 1mm slice thickness, clearly displaying the main intracranial anatomical structures.

[0026] Subsequently, CT perfusion imaging is performed: an iodine contrast agent (e.g., 50 mL, injected at a rate of approximately 5 mL / s) is administered via a peripheral vein, while a continuous dynamic scan of the target brain region is performed simultaneously for approximately 50 seconds, obtaining a series of time-series CT images. CTP typically covers multiple key planes of the brain (e.g., 8–10 layers, approximately 80 mm longitudinal range), with a temporal resolution of 1–2 seconds, thereby capturing the changes in brain tissue perfusion over time.

[0027] The acquired dynamic image data is input into perfusion analysis software, and algorithms such as deconvolution are used to calculate and generate cerebral blood flow perfusion parameter maps, including indicators such as cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT). CBF reflects the blood flow rate per 100g of brain tissue per unit time (unit: mL / 100g / min), CBV represents the blood volume contained in 100g of brain tissue (mL / 100g), and MTT is the average time (seconds) for blood to pass through the local capillary network of brain tissue. By analyzing these perfusion parameter maps, the ischemic penumbra and infarct core areas of brain tissue can be identified, providing important information for subsequent determination of the location of thrombus occlusion.

[0028] In addition, this module can also acquire cerebral angiography image data to construct a cerebral vascular anatomy model. Preferably, CT angiography (CTA) is performed after CTP scanning: that is, using arterial phase scanning after contrast agent injection to quickly acquire three-dimensional CTA image data displaying the cerebral arterial tree. Alternatively, magnetic resonance angiography (MRA) sequences can be used to acquire cranial vascular images. The acquired CTA or MRA data will be provided to the subsequent image registration and vascular modeling module for reconstructing the cerebral vascular structure model.

[0029] Image registration and brain blood vessel modeling

[0030] The image registration and cerebral vascular modeling module is used to spatially align and fuse multimodal medical images and reconstruct a three-dimensional structural model of intracranial blood vessels. In the registration processing of CT and CTP images, this invention employs an elastic registration method to correct positional differences between different imaging modes. Preferably, the B-spline free deformation (B-splineFFD) algorithm is used to achieve high-precision alignment between CT plain scan images and CTP parametric images. First, rigid registration correction is performed on the multi-phase sequence images of CTP to eliminate possible motion artifacts during patient scanning; then, a specific phase (or a comprehensive image obtained by weighted averaging and fusing multiple phase images) is selected as a floating image and registered to the reference CT plain scan image. During the multi-resolution registration optimization process, mutual information (MI) is used as a multimodal image similarity metric, and the position of the B-spline control grid nodes is iteratively optimized to maximize the overlap between CT anatomical structures and CTP functional images.

[0031] In another embodiment, a feature-based registration algorithm can also be used to fuse CT and CTP images. For example, shape features of the skull contour, brain tissue boundaries, or major vascular structures in CT and CTP images can be extracted, and the deformation field can be calculated to align the CTP parametric map to the CT image. By combining shape matching with rigid body transformation for initial alignment, and then applying local non-rigid transformation for fine-tuning, corresponding regions of the two images can be accurately overlapped. Once the above registration process is completed, the CT structural image and the CTP perfusion image are in a unified coordinate system, laying the foundation for subsequent fusion analysis.

[0032] For registered CT and CTP data, appropriate image fusion strategies can be employed to integrate anatomical and functional information. Specifically, the grayscale anatomical image from CT is used as the base image, and a semi-transparent overlay of CTP perfusion parameter maps (such as color-coded parameters like CBF and MTT) is placed on top, simultaneously presenting brain tissue structure and perfusion status within the same image. With the fused image, physicians can visually observe the positional correspondence between areas of abnormal perfusion and anatomical structures. For example, when a region exhibits significantly reduced CBF and markedly prolonged MTT, this region will be highlighted in the fusion result, indicating potential stenosis or blockage in its supplying vessels. Image fusion helps to accurately locate the extent of ischemic areas and provides a basis for diagnosing related vascular lesions.

[0033] This module further segments and models the acquired cerebral angiography images (CTA or MRA) in 3D to reconstruct the structural atlas of cerebral blood vessels. Preferably, a deep learning algorithm is used for automatic vessel segmentation. For example, a U-Net convolutional neural network model is trained, taking the CTA 3D image data as input for voxel-level segmentation and outputting a probability distribution map of the vessels. Subsequently, a threshold segmentation is applied to the probability map to extract a binary mask of the 3D cerebral blood vessels. Alternatively, an encoder-decoder neural network model such as SegNet can be used to achieve similar vessel segmentation results. To improve the accuracy of the vessel model, morphological post-processing can be performed on the initial segmentation results, including removing isolated small noise points, filling small holes, and extracting fine center lines of the vessels to ensure vessel connectivity. After these steps, a complete and coherent 3D model of cerebral blood vessels can be obtained, clearly presenting the structure of the intracranial vascular tree.

[0034] Based on this, a topological model of the cerebral vascular network is constructed. By extracting the centerline and bifurcation nodes from the 3D vascular model, the vascular structure can be abstracted as a connected graph in node-edge form: the bifurcation points and terminal points of each major blood vessel serve as nodes in the graph, and a continuous segment of blood vessel between adjacent bifurcation points corresponds to an edge. Each edge inherits the geometric attributes (such as path length and average diameter) and connectivity (upstream and downstream nodes) of the corresponding vascular segment. The resulting vascular topology data not only preserves the anatomical morphology of cerebral blood vessels but also represents the connectivity of blood vessels in a graph structure, which will serve as the input to the path planning module. Under the effect of fusion registration, the vascular topology model extracted by CTA / MRA and the perfusion parameter information provided by CTP are in a unified coordinate system. This integrated model clearly shows the spatial location of each vascular segment in the brain and the perfusion status of its blood supply area, which can be used to determine the target vascular branch where a suspected thrombus is located, providing a reliable basis for subsequent path planning and navigation of the microrobot.

[0035] Path planning

[0036] The path planning module utilizes the aforementioned cerebral vascular topology model to calculate the optimal route for the microrobot from entering the blood vessel to reaching the target thrombus region. First, the cerebral vascular network is abstractly modeled as a directed graph structure, where vascular bifurcation nodes, terminal nodes, and the starting point of the entry route (such as the internal carotid artery entrance) serve as vertices, and vascular segments serve as directed edges connecting the vertices. The direction of the edges is set according to the physiological direction of blood flow from the proximal end to the distal tissue. This directed graph accurately reflects the connection relationships and blood flow direction of the cerebral artery vascular tree and can be used to search for pathways from external entry points to the target occlusion point.

[0037] The path search employs the A algorithm to find the optimal path on the aforementioned weighted directed graph. A is a heuristic shortest path search algorithm that expands nodes based on both the actual and estimated costs of the path, efficiently finding a low-cost path from the starting point to the destination. In this embodiment, the entry node where the microrobot enters the blood vessel is taken as the starting point, and the terminal node corresponding to the target blood vessel segment where the embolism occurs is taken as the target node. The algorithm maintains a priority queue to store nodes to be explored and calculates an evaluation function f(n) = g(n) + h(n) for each node: where g(n) is the cumulative actual cost from the starting point along the traversed path to node n, and h(n) is the heuristically estimated cost from node n to the target. In each iteration, the node with the smallest f(n) is selected for expansion, gradually advancing the search until the target node is reached, obtaining candidate path solutions.

[0038] To reasonably evaluate path costs, it is necessary to design weight functions for the edges in the graph that incorporate factors such as vessel size and perfusion status. Generally, smaller vessel radii mean greater resistance and higher risk for the microrobot to pass through, thus requiring a higher cost for that path segment. Conversely, if a vessel supply area has severe perfusion anomalies (suggesting potential stenosis or even occlusion), the path cost through that segment will also increase accordingly. The cost of an edge can be set as a functional combination of vessel length, diameter, and perfusion factor. For example, the cost can be directly proportional to vessel length and inversely proportional to vessel diameter, with an added penalty factor related to the degree of perfusion anomaly. For instance, the following cost function could be used: Where L is the length of the vessel segment, D is its average diameter, E represents the degree of perfusion abnormality in the area supplied by the segment (normalized to between 0 and 1, with 0 for normal and 1 for complete ischemia), and k is a weighting coefficient that determines the influence weight of perfusion factors. This definition makes vessel segments with smaller diameters or severe ischemia in the supplied area more costly, thus guiding the algorithm to prioritize paths with wider diameters and better blood flow.

[0039] The choice of the heuristic function h(n) is based on an optimistic estimate of the remaining path cost. In this embodiment, the heuristic value can be simply taken as the straight-line Euclidean distance between the current node position and the target node position (i.e., the spatial distance between the two points), and appropriately divided by an average vascular tortuosity coefficient or an assumed average velocity constant to ensure unit consistency. Since the Euclidean straight-line distance is always less than or equal to the actual distance along the vascular pathway, this heuristic function guarantees the underestimation of the remaining cost (satisfying the consistency requirement of the A algorithm), thereby ensuring the accuracy of the search process and the availability of the optimal solution.

[0040] Once Algorithm A finds the path to the target, it outputs the optimal path and its specific route information. The path can be represented as a sequence of nodes from the starting node through several intermediate vascular nodes to the target node, or equivalently as a list of vascular segments traversed sequentially. The system can further map this discrete path into a continuous three-dimensional spatial trajectory: for example, by extracting interpolated path points between nodes based on the coordinates of the vascular centerline, a smooth three-dimensional curve trajectory is formed, representing the route the microrobot must take along the blood vessel. This trajectory can be visually displayed on three-dimensional images before surgery for doctors to verify its rationality. Once the path is determined, the subsequent magnetic navigation control module of the microrobot will use this trajectory as a reference to guide the robot segment by segment to reach the thrombus location. If the original planned path is blocked due to unforeseen circumstances during navigation, the system can also replan a new path in real time, ensuring the robot can bypass obstacles and continue towards the target.

[0041] Magnetic navigation control for micro robots

[0042] A microrobot magnetic navigation control module is used to manipulate a micro thrombolysis robot to move along a planned path within a vascular network. The microrobot has a miniaturized structure and built-in magnetic components, enabling it to achieve directional movement in response to an external magnetic field. In a preferred embodiment, the microrobot has a diameter on the sub-millimeter scale (hundreds of micrometers), is generally elongated cylindrical or spherical, and has a small segment of strongly magnetic material (such as a neodymium iron boron micromagnetic rod) encapsulated inside. This magnetic material imparts a fixed magnetic moment to the robot (e.g., the direction of the magnetic moment is aligned with the robot's long axis). The robot's exterior is coated with a biocompatible material to prevent blood from contacting metal ions, improving safety. This robot design is large enough to pass through narrow segments of cerebral blood vessels without damaging the vessel walls, providing the hardware basis for in vivo magnetic navigation.

[0043] This system wirelessly drives and navigates a microrobot using a controllable magnetic field generated externally. The magnetic navigation device consists of multiple sets of electromagnetic coils distributed around the patient's head, generating a three-dimensionally adjustable magnetic field and its gradient in the target area. When the coils are energized, a magnetic field with a magnetic induction intensity of B is established in the space where the robot is located. The magnetic moment m inside the microrobot will be subjected to force and torque in the external magnetic field: on the one hand, the magnetic dipole is subjected to torque. This causes the robot's magnetic moment direction to quickly align with the direction of the external magnetic field, thereby enabling control over the robot's orientation; on the other hand, in the presence of a gradient in a non-uniform magnetic field, the robot experiences a net magnetic force. The robot's magnetic field tends to pull it towards areas with stronger magnetic fields, providing the propulsion for its movement. By coordinating and controlling the direction and spatial gradient of the magnetic field, remote, synchronized control of the robot's posture and motion can be achieved.

[0044] To accurately describe the force-motion relationship of the robot in a magnetic field, a state-space dynamics model of the microrobot is established. The robot's position and velocity are taken as the system states. Where p=(x,y,z) are the robot's position coordinates in three-dimensional space, and v=(v x ,v y ,v z Let be its velocity vector. The kinematic characteristics of the robot in the blood environment are dominated by viscous drag, which can be expressed by the second-order dynamic equations as follows:

[0045] m r =F mag -F drag ,d p \d t =v,

[0046] Where m r For the quality of the robot, F magLet F be the driving thrust generated by the magnetic field gradient, and F be the viscous drag of the blood on the robot (which can be approximated as γv, where γ is a constant damping coefficient). The magnetic driving force Fmag can be expressed as: F map (p,t)=∇(m·B(p,t)). Its direction is along the direction of increasing magnetic field gradient, and its magnitude is approximately proportional to the product of magnetic moment intensity and magnetic field gradient. In the low Reynolds number fluid environment within blood vessels, the inertial term has negligible influence, and the acceleration term can be ignored to simplify the system into a quasi-static model, i.e., assuming instantaneous force equilibrium: F mag +F. The above dynamic model characterizes the position and velocity evolution of the microrobot under the influence of a magnetic field, providing a theoretical basis for control system design.

[0047] v≈(1 / у)˙F mag Based on the above model, a magnetic field gradient navigation control strategy is formulated to guide the robot along the planned path. The control input is the magnetic field vector and its gradient (B, ∇B) generated by the external coil. The control strategy is as follows: the direction of magnetic field B is always oriented towards the line connecting the robot's current position to the next target path point, guiding the robot's magnetic moment to align with the desired direction; simultaneously, a magnetic field strength gradient ‖∇B‖ of a certain magnitude is applied in this direction, generating a net magnetic force that pulls the robot towards the target direction. By continuously adjusting the magnetic field direction along the path, the microrobot can achieve smooth turning at blood vessel bifurcation points and maintain centered movement along the blood vessel axis on straight sections. The magnetic field gradient strength is adjusted in real time according to the navigation phase: the gradient is increased to provide greater traction and increase the robot's speed when acceleration is needed; while the gradient strength is appropriately reduced when approaching the target area to decrease speed and facilitate precise stopping of the robot. Using the above magnetic field gradient navigation strategy, the microrobot can overcome the influence of blood flow, move stably along the predetermined three-dimensional blood vessel path, and ultimately accurately reach the blocked thrombus site for targeted thrombolysis.

[0048] Closed-loop feedback control system design

[0049] The closed-loop feedback control module is used to realize the automatic control and real-time correction of micro-robot navigation, including real-time robot positioning, trajectory tracking control algorithms, and magnetic field control execution. Through a sensing and control closed loop, the system ensures that the robot travels along the planned path and can correct its course promptly when disturbed.

[0050] Real-time localization method (magnetic + image fusion): This system combines magnetic field sensing and medical imaging to monitor the position of the microrobot in real time. First, a multi-channel magnetic sensor array is deployed around the patient's head to collect weak magnetic field signals generated by the microrobot's built-in magnetic moment in real time. The three-dimensional coordinates of the robot are then calculated using a differential calculation algorithm based on the multi-point magnetic signals. Simultaneously, medical imaging is used to intermittently verify the robot's position. For example, when necessary, digital subtraction angiography (DSA) is used to obtain fluoroscopic images of the robot's position within blood vessels, or a fast MRI sequence is used for localization scanning. Fusing the position estimate calculated by the magnetic sensors with the position measured by the images (e.g., using a Kalman filter fusion algorithm) overcomes the limitations of a single method, improving localization accuracy and robustness. With magnetic-image fusion localization, the control system can continuously obtain the accurate position of the microrobot within the brain, providing real-time feedback for closed-loop control.

[0051] The relationship between magnetic field control input and target deviation: The controller adjusts the magnetic field input in real time based on the deviation between the current state and the target state of the microrobot. The desired trajectory is obtained from path planning, and the positional difference between the current robot and the desired trajectory is defined as the deviation vector e=p. ref -p (including position and orientation errors). The control system uses this deviation as a basis to calculate the required magnetic field correction commands: for example, when the robot has a lateral deviation, the magnetic field direction is appropriately rotated towards the direction of deviation to correct the heading deviation; when the robot is still far from the target point, the magnetic field gradient is increased to accelerate the forward speed, while the gradient intensity is reduced when approaching the target to decelerate and accurately position the robot. Through this control method that uses error as a feedback signal, if the robot deviates during movement, it can automatically correct its direction and approach the desired path; as it gradually approaches the target point, the controller will smoothly reduce the magnetic field traction force, so that the robot decelerates smoothly and stops accurately at the target position, preventing it from overshooting the target or colliding with the blood vessel wall.

[0052] Objective Function Design (Error + Energy Consumption Trade-off): To optimize control performance while considering safety and energy consumption, a performance objective function needs to be designed for the navigation controller. This objective function typically includes a comprehensive measure of tracking error and control energy consumption, guiding the optimization of the control law. The error term can be defined as the sum of squares (or integral over time) of the deviations between the robot's actual trajectory and the reference path, reflecting the navigation accuracy requirements; the energy consumption term can be defined as the sum of squares of the amplitude or rate of change of the magnetic field control input, reflecting the intensity and stability requirements of the applied magnetic field. Taking continuous-time form as an example, the performance index function can be set as follows:

[0053] ,

[0054] Where e(t) is the position error state vector, u(t) is the magnetic control input vector (e.g., coil current or equivalent magnetic field parameters), and matrices Q and R are positive definite weight matrices, corresponding to the weighting coefficients of the error term and energy consumption term, respectively. By reasonably selecting the weights of each state component in Q (e.g., assigning higher weights to position deviations and appropriate weights to velocity), and setting a suitable R value to limit the control input amplitude (e.g., avoiding energy consumption and safety hazards caused by excessive magnetic field strength), the magnetic field energy output can be controlled within a reasonable range while ensuring navigation accuracy. This objective function provides an evaluation basis for the optimization of subsequent control strategies (such as LQR or MPC).

[0055] LQR Control Strategy Design: Based on the above performance indicators, a linear quadratic regulator (LQR) can be used to optimize the closed-loop control of the linearized model of the microrobot. First, the dynamic model of the microrobot's magnetic navigation is linearized near the operating point, resulting in the form x=A x +B u The state-space equation is given, where x is the robot's deviation state vector from the desired trajectory, and u is the coil magnetic field control input vector. For the linear model, a quadratic objective function is introduced. The optimal feedback gain matrix K is obtained by solving the Riccati equation. Based on this, a linear state feedback control law u is constructed. (t) =-Kx (t) That is, based on the current state deviation x (t) By linearly weighting the gain matrix K, the input quantity of the correction magnetic field (e.g., the increase or decrease of the current in each coil) is calculated in real time to correct the robot's trajectory deviation. The LQR controller has a simple feedback structure and fast calculation, ensuring the asymptotic stability of the closed-loop system under ideal conditions and minimizing the aforementioned performance index J. In terms of parameter design, the diagonal elements related to position error in the Q matrix are usually given larger weights to strictly constrain the magnitude of the robot's deviation from the path; while the value of the R matrix is ​​relatively increased to increase the penalty for control input overhead (to avoid excessive coil current or excessive magnetic field strength). By selecting appropriate Q and R matrix values ​​through simulation and debugging, control energy consumption and magnetic field effects can be limited within a safe range while ensuring rapid response and small error.

[0056] MPC Control Strategy Design: In another implementation, a Model Predictive Control (MPC) strategy is used to achieve closed-loop navigation. MPC improves the accuracy and robustness of trajectory tracking while considering system constraints by optimizing the control sequence within a finite prediction time domain. Specifically, it is based on a discrete-time dynamics model of the robot. We select a prediction time domain length of N steps and a control time domain length of M steps. We construct a quadratic cost function similar to LQR:

[0057] ,

[0058] And add necessary constraints (such as limiting the maximum value and rate of change of the magnetic field strength to avoid adverse effects on the patient). At each control moment, with the current state x [k] Initially, solve the optimization problem Jk online to obtain the optimal control sequence {u} for the next M steps. [k] , ...,u [k+M-1] In actual execution, only the first step of this sequence, control u, is applied. [k] =u [k] The system then enters the next time step k+1 to re-acquire the state and perform rolling optimization, repeating this cycle to achieve real-time feedback control. Compared to LQR, the MPC method can explicitly handle control inputs and state constraints, and its rolling optimization of the predictive model provides foresight regarding potential future path changes or disturbances. Regarding parameter selection, the prediction time domain length N should be sufficient to cover the time required for the robot to adjust back to the path from its current deviation; for example, if it is estimated that the robot takes approximately 2 seconds to complete a 90-degree turn and deceleration, and the control cycle is 0.1 seconds, N can be set to approximately 20 steps. The control time domain length M can be equal to or slightly shorter than N to reduce the computational load. The selection principles for the weight matrices Q and R are similar to those of LQR; adjusting their values ​​allows for a trade-off between response speed and control stability. Since MPC requires solving the optimization problem online in each cycle, the control system must have sufficient computational power or employ a fast optimization algorithm (such as using a quadratic programming QP solver) to ensure that the control update frequency meets the real-time control requirements.

[0059] In summary, the closed-loop feedback control system achieves precise navigation control of the microrobot through high-precision real-time positioning, advanced control algorithms (LQR / MPC), and a well-designed objective function. The system continuously adjusts the magnetic field direction and gradient based on the deviation between the robot and the target path, enabling automatic guidance and dynamic correction. When the robot successfully reaches the target thrombus location along the planned route, the control system can stop it and perform local thrombolysis. Throughout the process, the closed-loop control strategy ensures the robot's safe and efficient movement, enabling automated implementation of microrobot-guided targeted thrombolysis in complex cerebrovascular environments.

Claims

1. A CT-CTP registration guided magnetic micro-robotic directional thrombolytic system, comprising: The CT / CTP image acquisition module, the image registration and cerebral vessel modeling module, the path planning module, the micro-robot magnetic navigation control module, and the closed-loop feedback control module are characterized in that the CT / CTP image acquisition module is used to acquire tomographic anatomical images and perfusion functional images of the brain of a patient; the image registration and cerebral vessel modeling module performs high-precision registration and fusion of the tomographic anatomical images and the perfusion functional images, performs spatial alignment of CT images and CTP perfusion parameter images through a multi-resolution free-form deformation model, takes mutual information as a registration similarity measure, corrects pose differences between different modal images, makes CT anatomical structures and CTP functional images coincide in a unified coordinate system, and reconstructs a three-dimensional structural model of cerebral vessels based on acquired cerebral vessel image data; the path planning module plans a travel path of a micro-robot from a vessel entrance to a target thrombus position by using the three-dimensional model of cerebral vessels and perfusion information; the micro-robot magnetic navigation control module drives the micro-robot to move along the travel path through a controllable magnetic field generated externally; and the closed-loop feedback control module monitors the position of the micro-robot in real time and automatically adjusts magnetic field parameters according to deviations to guide the micro-robot to accurately navigate to a thrombus site to implement site-specific thrombolysis.

2. The CT-CTP registration guided magnetic micro-robotic directional thrombolytic system according to claim 1, wherein, The CT / CTP image acquisition module adopts spiral CT scanning to acquire plain CT images and CT perfusion images of the head of a patient; first, non-enhanced CT structural images of the brain are acquired, then contrast agent is injected through peripheral veins and dynamic CT perfusion scanning is performed on the target region of the brain for about 50 seconds to obtain a series of time-series CT images and generate cerebral blood perfusion parameter maps, including cerebral blood flow, cerebral blood volume, and mean transit time, for identifying ischemic penumbra and infarct core regions of brain tissue, thereby providing basic data for determining the position of a thrombus blockage.

3. The CT-CTP registration guided magnetic micro-robotic directional thrombolytic system according to claim 2, wherein, After the CT perfusion imaging is completed, the image acquisition module further performs cerebral angiography imaging to acquire three-dimensional image data of cerebral vessels; the cerebral angiography images adopt CT angiography scanning to acquire three-dimensional CTA image data of cerebral arterial vessel trees, or adopt magnetic resonance angiography sequences to acquire cerebral vessel image data, which are used by the image registration and cerebral vessel modeling module to reconstruct an anatomical structural model of cerebral vessels.

4. The CT-CTP registration guided magnetic micro-robotic directional thrombolytic system according to claim 3, wherein, The image registration and cerebral vessel modeling module performs segmentation and three-dimensional reconstruction on the acquired CTA or MRA cerebral vessel images to obtain a three-dimensional structural model of cerebral vessel trees; the module uses a deep learning algorithm to automatically segment the cerebral vessel images, extracts a three-dimensional binary model of the vessels, and performs morphological processing and centerline extraction on the model to generate a continuous and connected cerebral vessel topology model that clearly presents the bifurcation structure of intracranial vessels and the geometric properties of each vessel segment, and combines the CTP perfusion parameter information to give the cerebral vessel topology model corresponding perfusion state data.

5. The CT-CTP registration guided magnetic micro-robotic directional thrombolytic system according to claim 4, wherein, The path planning module abstracts the cerebral vascular topology model as a directed graph structure, and plans an optimal path for the micro robot from an entrance node to a blood vessel node where the target thrombus is located on the directed graph using a heuristic search algorithm; an evaluation function f(n)=g(n)+h(n) is set during path planning, where g(n) is the cumulative path cost from the starting point to the current node n, and h(n) is the heuristic estimated cost from node n to the target; a weight cost function is designed for the edges of the vascular directed graph, which combines the length and diameter of the blood vessel segment, the degree of perfusion abnormality of the blood supply area, and other factors to calculate the path cost, so that the path cost is proportional to the length of the blood vessel and inversely proportional to the diameter of the blood vessel, and the penalty weight is increased for the blood vessel segment with severe perfusion abnormality, thereby guiding the algorithm to preferentially select the blood vessel path with larger diameter and normal blood supply; the heuristic search algorithm selects the node with the minimum evaluation function f(n) for iterative expansion until the optimal path to the target is found.

6. The CT-CTP registration guided magnetic micro-robotic directional thrombolytic system of claim 1, wherein, The micro robot magnetic navigation control module includes a micro thrombolytic robot and an external magnetic field generating device; the micro thrombolytic robot is sub-millimeter in size, in the shape of an elongated cylinder or a sphere, has a directional magnetic element sealed inside to generate a fixed direction magnetic moment, and is coated with a biocompatible material on the outer surface, so that it can safely pass through the narrow blood vessel segment in the blood vessel lumen and not damage the blood vessel inner wall; the external magnetic field generating device is composed of multiple groups of electromagnetic coils distributed around the patient's head, which is used to generate a three-dimensional adjustable magnetic field and a magnetic field gradient to wirelessly drive the motion of the micro robot and control its orientation, the change of the magnetic field direction is used to twist the internal magnetic moment of the robot to align the robot orientation with the predetermined path direction, and the magnetic force generated by the magnetic field gradient is used to pull the robot forward in the direction where the magnetic field strength increases.

7. The CT-CTP registration guided magnetic micro-robotic directional thrombolytic system according to claim 6, wherein, The external magnetic field generating device adjusts the posture of the micro robot in the vascular network and controls the propulsion force by coordinating the direction of the magnetic field vector and the gradient strength; the direction of the magnetic field is always directed to the direction of the next target path point from the current position of the micro robot, so as to guide the alignment of the magnetic moment of the robot with the expected heading, and the gradient strength of the magnetic field is adjusted in real time according to the navigation stage: the gradient is increased to provide greater traction when the robot needs to accelerate, and the gradient strength is reduced to slow down when approaching the target thrombus area for accurate positioning, thereby enabling the micro robot to smoothly turn, center and accurately stop at the target thrombus for directional thrombolysis along the blood vessel path.

8. The CT-CTP registration guided magnetic micro-robotic directional thrombolytic system of claim 6, wherein, The closed-loop feedback control module comprises a micro robot real-time positioning unit for monitoring the position of the micro robot in combination with magnetic signal sensing and medical imaging; the real-time positioning unit acquires weak magnetic field signals generated by the built-in magnetic moment of the micro robot in real time through a multi-channel magnetic field sensor array arranged around the head of the patient, and calculates the spatial coordinate position of the robot, while using medical imaging equipment to check the position of the robot in the blood vessel, wherein digital subtraction angiography (DSA) can be used to obtain the perspective position image of the robot, or fast MRI sequence can be used to position the robot; the magnetic sensing positioning result and the imaging positioning result are fused and corrected through a data fusion algorithm, and Kalman filtering is used to realize the fusion of the sensing data, so as to improve the accuracy and robustness of the micro robot positioning, and provide high-precision position feedback signals for the closed-loop feedback control module.

9. The CT-CTP registration guided magnetic micro robot directed thrombolytic system according to claim 8, wherein, The closed-loop feedback control module comprises a trajectory tracking control unit for automatically adjusting the magnetic field control input according to the current position deviation of the micro robot to make the robot travel along the predetermined path; the trajectory tracking control unit obtains the expected trajectory of the planned path and the current position of the robot, calculates the deviation vector of the robot, including the position deviation and the orientation deviation, and inputs the deviation into the control algorithm to generate the magnetic field correction instruction; the control algorithm uses a pre-designed performance index function to balance the trajectory tracking error and the control energy consumption, and correspondingly uses a linear quadratic regulator control strategy or a model predictive control strategy to perform feedback control on the robot: when the linear quadratic regulator control strategy is used for control, the magnetic navigation system of the micro robot is linearly modeled and the optimal state feedback gain matrix is solved based on the performance index, and the magnetic coil current adjustment amount is calculated in real time to correct the position deviation of the robot; when the model predictive control strategy is used for control, the magnetic field control sequence is optimized within a limited prediction time domain based on the kinematics model of the robot, and the constraints of magnetic field strength and change rate are explicitly considered, so as to improve the accuracy, stability and robustness of trajectory tracking, and ensure that the micro robot safely and efficiently reaches the target position and implements thrombolysis in the complex cerebral vascular environment.

Images