Lumbar pedicle screw dynamic guiding method and system based on augmented reality

Abstract

The invention relates to the technical field of intelligent medical treatment, in particular to a lumbar vertebral pedicle screw dynamic guiding method and system based on augmented reality, and the method comprises the following steps: constructing an entity lumbar vertebra skeleton model of a patient, and obtaining a cubic marker of the entity lumbar vertebra skeleton model to establish a lumbar vertebra skeleton space coordinate system; acquiring scanning data of the entity lumbar vertebra skeleton model based on the lumbar vertebra skeleton space coordinate system, and reconstructing a three-dimensional lumbar vertebra skeleton model according to the scanning data; combining the entity lumbar vertebra skeleton model and the three-dimensional lumbar vertebra skeleton model to obtain a lumbar vertebra skeleton virtual and real registration result; and obtaining a real-time path track of the lumbar vertebra pedicle screw according to the lumbar vertebra skeleton virtual and real registration result so as to realize dynamic guidance of the lumbar vertebra pedicle screw. According to the invention, through registration of the solid model and the three-dimensional model, dynamic guidance of the screw path is realized in combination with a risk early warning mechanism, and the accuracy and safety of implantation of the lumbar pedicle screw are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, and in particular to a dynamic guidance method and system for lumbar pedicle screws based on augmented reality. Background Technology

[0002] In lumbar pedicle screw implantation surgery, current technologies primarily rely on a combination of preoperative static imaging and intraoperative optical or electromagnetic tracking systems. A typical procedure involves: preoperatively acquiring three-dimensional images of the patient's lumbar spine via scanning equipment, and the surgeon planning the screw path on the images; intraoperatively, optical markers or electromagnetic sensors are fixed to the patient's bones or skin, and the position of surgical instruments is tracked in real time using an infrared camera or electromagnetic field positioning device. The preoperative images are then registered with the actual intraoperative anatomical structures, and the planned path is finally overlaid on the surgeon's view in augmented reality. Some technologies employ mechanical guides or robot assistance, limiting the instrument's range of motion through pre-set mechanical constraints. The core of this approach lies in constructing a "static digital twin" through preoperative imaging, combined with intraoperative positioning to achieve a mapping between the "virtual path" and the "actual operation," essentially representing an offline "plan first, execute later" model.

[0003] Existing technologies have the following drawbacks: First, the preoperative 3D model and the patient's actual intraoperative anatomical structure are prone to registration deviations due to changes in body position and tissue deformation. In particular, surface markers are easily affected by soft tissue movement, leading to decreased navigation accuracy. Second, intraoperative fluoroscopic images are two-dimensional planar information, lacking three-dimensional spatial relationships, making it difficult to quantify the distance between the screw and the pedicle cortex and the risk of perforation in real time. Third, they rely heavily on fixation equipment and lack a real-time compensation mechanism, making it impossible to dynamically adapt to registration errors caused by real-time changes in the position and posture of surgical instruments. Fourth, traditional path planning uses a fixed angle, ignoring individual differences in pedicle curvature and lacking real-time monitoring and feedback of parameters such as pressure and speed during screw implantation, increasing the risk of surgical complications. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a dynamic guidance method and system for lumbar pedicle screws based on augmented reality.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a dynamic guidance method for lumbar pedicle screws based on augmented reality. The method includes the following steps: constructing a physical lumbar skeletal model of the patient and acquiring cubic markers from the physical lumbar skeletal model to establish a spatial coordinate system for the lumbar skeletal system; acquiring scanning data of the physical lumbar skeletal model based on the spatial coordinate system, and reconstructing a three-dimensional lumbar skeletal model based on the scanning data; obtaining a virtual-to-real registration result of the lumbar skeletal system by combining the physical lumbar skeletal model and the three-dimensional lumbar skeletal model; and obtaining the real-time path trajectory of the lumbar pedicle screw based on the virtual-to-real registration result, thereby achieving dynamic guidance of the lumbar pedicle screw. The present invention improves registration accuracy by accurately registering the physical model and the three-dimensional model, establishing a stable spatial coordinate system by combining cubic markers, and dynamically acquiring the real-time path trajectory of the screw, thereby avoiding the risk of perforation in real time and improving the safety and precision of the surgery.

[0006] Optionally, constructing a physical lumbar spine model of the patient and obtaining cubic markers from the physical lumbar spine model to establish a spatial coordinate system for the lumbar spine includes: obtaining the patient's lumbar spine structure; fabricating the physical lumbar spine model based on the lumbar spine structure using 3D printing technology; fixing the cubic markers on the iliac crest of the physical lumbar spine model and embedding near-infrared reflective coatings at the edges of the cubic markers; and using the geometric center of the cubic markers as a spatial positioning reference point to establish the spatial coordinate system for the lumbar spine. This invention recreates the patient's lumbar spine structure and fixes cubic markers at the iliac crest, establishing a coordinate system using their geometric center, thereby improving the realism of the physical model, enhancing the stability of marker recognition, and providing a precise spatial reference for subsequent registration.

[0007] Optionally, the step of acquiring scanning data of the physical lumbar spine model based on the lumbar spine spatial coordinate system and reconstructing a three-dimensional lumbar spine model based on the scanning data includes: dynamically adjusting the X-ray scanning angle based on the direction of the lumbar spine spatial coordinate system to acquire the scanning data; segmenting the scanning data to extract the pedicle cortical boundary of the physical lumbar spine model, using the pedicle cortical boundary as the segmentation result; acquiring curvature constraints, and reconstructing the three-dimensional lumbar spine model by combining the segmentation result and the curvature constraints. This invention dynamically adjusts the X-ray angle based on the coordinate system to acquire scanning data, accurately extracts the pedicle cortical boundary, and reconstructs a three-dimensional model by combining curvature constraints, improving scanning adaptability and segmentation accuracy, ensuring the model fits the physical anatomical structure, and laying the foundation for subsequent virtual-real registration.

[0008] Optionally, obtaining the curvature constraint includes: extracting the pedicle centerline based on the segmentation result, and obtaining the discrete curvature of the pedicle centerline to construct a pedicle curvature field; constructing a curvature-allowable error mapping function based on the pedicle curvature field, and using the curvature-allowable error mapping function as the curvature constraint. This invention, by extracting the discrete curvature of the pedicle centerline to construct a curvature field and establishing a curvature-allowable error mapping function as a curvature constraint, accurately reflects the anatomical features of the pedicle, dynamically adjusts the model reconstruction error, and improves the morphological realism and structural accuracy of the three-dimensional model.

[0009] Optionally, obtaining the lumbar spine registration result by combining the physical lumbar spine model and the three-dimensional lumbar spine model includes: performing spatial registration between the physical lumbar spine model and the three-dimensional lumbar spine model to obtain an initial registration result for the lumbar spine; and dynamically correcting the initial registration result to obtain the final virtual-real registration result for the lumbar spine. This invention first obtains an initial result through spatial registration, and then dynamically corrects and optimizes the initial result, reducing the registration deviation between the physical model and the three-dimensional model, improving the accuracy of virtual-real mapping, ensuring real-time correspondence between their spatial positions, and providing a reliable registration basis for subsequent screw path planning.

[0010] Optionally, the step of spatially registering the solid lumbar vertebral skeleton model and the three-dimensional lumbar vertebral skeleton model to obtain the initial registration result of the lumbar vertebral skeleton includes: performing multimodal recognition on the cubic marker to obtain near-infrared features of its edges and geometric contour features to construct a composite feature point set; establishing a virtual-real registration state equation based on the composite feature point set; obtaining a virtual-real registration matrix based on the virtual-real registration state equation; constructing a deformation displacement field of the cubic marker based on local elastic deformation; and updating the virtual-real registration matrix based on the deformation displacement field to obtain the initial registration result of the lumbar vertebral skeleton. This invention constructs composite feature points through multimodal recognition, obtains the registration matrix by combining it with the state equation, and then updates the registration matrix using the deformation displacement field. This effectively reduces the influence of marker deformation, improves the accuracy and stability of the initial registration, and provides a high-quality data foundation for subsequent correction.

[0011] Optionally, the step of dynamically correcting the initial registration result of the lumbar vertebrae to obtain the virtual-real registration result of the lumbar vertebrae includes: constructing a dynamic correction engine; obtaining the pose drift vector of the initial registration result of the lumbar vertebrae based on the dynamic correction engine; constructing a pose feature backtracking equation based on the pose drift vector; obtaining a corrected registration matrix by combining the initial registration result of the lumbar vertebrae; obtaining the virtual-real overlap error based on the composite feature point set; and performing negative feedback optimization on the corrected registration matrix based on the virtual-real overlap error to obtain the virtual-real registration result of the lumbar vertebrae. This invention captures pose drift through a dynamic correction engine, generates a correction matrix by combining it with a backtracking equation, and then uses the virtual-real overlap error for negative feedback optimization to correct registration deviations in real time, significantly improving the stability and accuracy of virtual-real registration.

[0012] Optionally, obtaining the real-time path trajectory of the lumbar pedicle screw based on the lumbar vertebral registration result includes: obtaining pedicle channel parameters based on the three-dimensional lumbar vertebral model, the pedicle channel parameters including the radius of curvature of the pedicle axis and the cortical bone thickness; obtaining the insertion angle of the lumbar pedicle screw by combining the lumbar vertebral registration result and the pedicle channel parameters; obtaining the surgical instrument pose; obtaining the instrument-planned path deviation angle based on the surgical instrument pose; and dynamically updating the insertion angle according to the instrument-planned path deviation angle to obtain the real-time path trajectory. This invention combines pedicle channel parameters and lumbar registration results to determine the insertion angle, and dynamically updates the path through instrument pose deviation, achieving real-time adjustment of the screw trajectory, conforming to individual anatomical characteristics, and improving the accuracy of path planning and surgical adaptability.

[0013] Optionally, the dynamic guidance of the lumbar pedicle screw includes: obtaining the minimum distance between the screw axis and the cortical bone based on the real-time path trajectory; acquiring the screw insertion pressure value; adjusting the speed of the lumbar pedicle screw based on the screw insertion pressure value to obtain the screw insertion speed; establishing a real-time perforation risk warning mechanism based on the minimum distance between the screw axis and the cortical bone and the screw insertion speed; and dynamically guiding the lumbar pedicle screw according to the real-time perforation risk warning mechanism. This invention, by calculating the minimum distance between the screw and the cortical bone, combining the screw insertion pressure with speed adjustment, establishes a perforation risk warning mechanism, avoids perforation risks in real time, and dynamically guides screw implantation, effectively improving surgical safety and operational accuracy.

[0014] Secondly, this invention provides an augmented reality-based dynamic guidance system for lumbar pedicle screws. The system executes the augmented reality-based dynamic guidance method for lumbar pedicle screws provided by this invention. The system includes an input device, an output device, a processor, and a memory, all interconnected. The memory stores a computer program, which includes program instructions, and the processor is configured to invoke these instructions. This invention utilizes high-performance input / output devices, a processor, and a memory to collaboratively execute the dynamic guidance method, efficiently store and invoke program instructions, achieve inter-device linkage, ensure a smooth surgical guidance process, and improve system stability and ease of operation. Attached Figure Description

[0015] Figure 1 This is a flowchart of a dynamic guidance method for lumbar pedicle screws based on augmented reality, according to an embodiment of the present invention. Figure 2 This is a framework diagram of a dynamic guidance system for lumbar pedicle screws based on augmented reality, according to an embodiment of the present invention. Detailed Implementation

[0016] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0017] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0018] Please see Figure 1 One embodiment of the present invention provides a dynamic guidance method for lumbar pedicle screws based on augmented reality, the method comprising the following steps: S1. Construct a physical lumbar spine model of the patient and obtain the cubic markers of the physical lumbar spine model to establish a spatial coordinate system for the lumbar spine.

[0019] In this embodiment, the patient's lumbar vertebral bone structure is obtained, and a solid lumbar vertebral bone model is made based on the lumbar vertebral bone structure using 3D printing technology. A cube marker is fixed on the iliac crest of the solid lumbar vertebral bone model, and a near-infrared reflective coating is embedded in the corners of the cube marker. The geometric center of the cube marker is used as a spatial positioning reference point to establish a spatial coordinate system for the lumbar vertebral bone.

[0020] In the preparation of the physical lumbar spine model, high-resolution lumbar spine data from the patient's preoperative computed tomography (CT) scans were used to separate the digital lumbar spine structure using medical image processing software. The lumbar spine structure was then imported into a photopolymerization 3D printer, and a biocompatible photosensitive resin was used for layered fabrication. The layer thickness was set to 25 μm to reproduce the microporous structure of the trabecular bone. After ultrasonic cleaning and secondary curing, the printed physical lumbar spine model was verified for anatomical accuracy, ensuring that the pedicle diameter error was ≤0.1 mm and the vertebral body height deviation was ≤0.3 mm.

[0021] Subsequently, the physical lumbar spine model underwent tissue environment simulation encapsulation: the physical lumbar spine model was fixed in a custom mold, and hardened opaque agar gel was perfused layer by layer—the base layer consisted of a 5.0%±0.2% agar solution, incorporating 10% by volume silica microspheres (0.5mm-1.0mm in diameter, with an elastic modulus of 50kPa±5kPa to simulate the resilience of soft tissue), while titanium dioxide powder was added to control light transmittance. The mold temperature was maintained at 40℃ during perfusion to avoid thermal shock damage to the resin, and the thickness of each layer was controlled at 10mm. 1mm red silica tubing was pre-embedded between layers to construct a simulated vascular network. After complete perfusion, gradient cooling curing was performed: 50℃ for 3 hours, 25℃ for 5 hours, and then refrigerated at 4℃ for final setting. Ultimately, the resulting agar-bone complex needs to undergo mechanical and optical verification: mechanical verification uses a surgical drill to penetrate the agar layer with standard pressure (10N), measures the resistance value (target range 45N-55N) and observes the debris splash pattern to ensure similarity to real muscle tissue; optical verification uses an AR headset to test the recognition stability of markers under 50% agar occlusion, thereby constructing a surgical simulation environment that combines mechanical realism, visual occlusion, and operational interactivity.

[0022] In this embodiment, a curvature matching point (local curvature radius 5mm ± 0.5mm) was selected in the iliac crest region of the physical lumbar spine model, and a biodegradable magnesium alloy anchor (2mm in diameter, 35% of the cortical bone thickness) was implanted. A custom titanium alloy cubic marker (10mm ± 0.3mm in edge length) was fixed to the anchor using a magnetic adsorption base. Its edges were chamfered at 45° and embedded with a near-infrared reflective coating (wavelength 850nm, reflectivity > 95%). A binary coding matrix was laser-engraved on the surface of the cubic marker for dual-modal recognition of CT scanning and augmented reality (AR). After fixation, the cubic marker underwent micro-vibration testing (6-DOF vibration table, frequency 5Hz / amplitude 0.1mm) to verify a displacement error of < 0.05mm. The coating reflection intensity was detected by the infrared camera of an AR device (such as Apple Vision Pro) to ensure stable tracking even when the instrument obscured 50%.

[0023] Furthermore, a right-handed Cartesian coordinate system was established with the geometric center of the cubic marker as the origin: the Z-axis was perpendicular to the simulated operating table plane, the X-axis was parallel to the line connecting the sacral promontory, and the Y-axis was determined according to the right-hand rule. Spatial coordinates of the eight vertices of the cubic marker were collected using a coordinate measuring machine, and the direction vectors of each axis were calculated and written into the model metadata. A solid lumbar spine model encapsulated in hardened agar was placed on the operating table, and the marker encoding was scanned using an AR device. A rigid body transformation algorithm was used to map the physical coordinate system to the AR virtual space, completing the establishment of the lumbar spine spatial coordinate system. This coordinate system serves as the basic reference for subsequent scan registration, supporting the need for millimeter-level dynamic guidance during surgery.

[0024] S2. Obtain the scanning data of the physical lumbar spine model based on the lumbar spine spatial coordinate system, and reconstruct the three-dimensional lumbar spine model based on the scanning data.

[0025] In this embodiment, the scanning angle of X-rays is dynamically adjusted based on the direction of the lumbar spine spatial coordinate system to obtain scanning data; the scanning data is segmented to extract the cortical boundary of the pedicle bone of the solid lumbar spine model, and the cortical boundary of the pedicle bone is used as the segmentation result; curvature constraints are obtained, and the three-dimensional lumbar spine model is reconstructed by combining the segmentation result and the curvature constraints.

[0026] During the data acquisition phase, CT scan parameters are first adjusted based on the established spatial coordinate system of the lumbar spine. Using the geometric center of a titanium alloy cube marker fixed to the iliac crest as the origin and its coordinate axis as the reference, the CT gantry is intelligently rotated: by calculating the spatial angle between the pedicle axis and the scanning plane in real time, the gantry rotation angle is dynamically adjusted to ensure that the central axis of each pedicle channel is perpendicular to the scanning plane. A layered dose strategy is employed: a 0.3mm ultra-thin slice thickness high-resolution mode is used in the pedicle region, while a conventional 0.8mm slice thickness is used in non-critical areas. Simultaneously, a metal artifact suppression algorithm is used to handle artifacts caused by the titanium alloy marker. During the scan, the binary encoding matrix laser-engraved on the marker surface is automatically identified to acquire scan data.

[0027] Furthermore, based on the spatial coordinate system, the pedicle region is located, and a two-stage segmentation algorithm is adopted: the first stage performs spatial weighted threshold segmentation based on marker coordinates, setting gradient threshold weights with the center of the pedicle as the sphere (a sphere region with a radius of 10mm), and prioritizing the extraction of the high-brightness area of ​​the bone cortex; the second stage applies edge detection, performing sub-pixel contour tracking along the bone texture direction, focusing on enhancing the boundary continuity of easily missed areas such as the pedicle isthmus and the pedicle-vertebral body connection, in order to obtain the boundary of the pedicle bone cortex; subsequently, the segmentation result is filled with micro-holes by morphological closing operation, and the output is a binary mask image that preserves the topological structure.

[0028] In this embodiment, the core of reconstructing the 3D lumbar spine model is model construction under curvature constraints. The pedicle centerline is extracted from the segmentation mask, and a smooth curve is generated using cubic spline interpolation, with discrete curvature values ​​calculated. A dynamic accuracy control strategy is established based on the curvature distribution: a maximum allowable error of 0.05mm is set for high curvature regions, forcing local mesh subdivision; the allowable error is relaxed to 0.1mm-0.2mm for medium-low curvature regions, employing adaptive simplification. A curvature consistency constraint is added during the reconstruction process, applying penalty weights to mesh surfaces deviating from the target curvature. After iterative optimization, the 3D lumbar spine model is obtained. Finally, lightweight compression is performed: a 3mm protection radius is set around markers to prevent simplification; other regions are compressed using a quadratic error metric algorithm. The pedicle channel curvature error is verified to be ≤0.08mm, and the marker vertex coordinate offset is ≤0.03mm. The output is a format file supporting real-time rendering by AR devices.

[0029] In this embodiment, the pedicle centerline is extracted based on the segmentation results, and the discrete curvature of the pedicle centerline is obtained to construct the pedicle curvature field; a curvature-allowable error mapping function is constructed based on the pedicle curvature field, and the curvature-allowable error mapping function is used as the curvature constraint condition.

[0030] Specifically, the centerline is first extracted based on the segmentation results of the pedicle cortex. Morphological thinning is then performed on the binary mask image, progressively peeling away the outer voxels until a single-pixel-width skeleton structure is generated. The coordinates of continuous center points are traced along the skeleton path, and a smooth curve is fitted using cubic spline interpolation. A sequence of 300-500 discrete points is generated by resampling at 0.1mm intervals. Local discrete curvature is calculated at each sampling point: tangent vectors are constructed from adjacent points to determine the reciprocal of the radius of curvature. For key locations such as the pedicle inlet and isthmus turning point, the sampling density is increased by three times to ensure data integrity in areas of abrupt curvature change. Finally, a centerline point cloud with curvature labels is output, forming a pedicle curvature field covering the entire pedicle channel.

[0031] Furthermore, a curvature-permissible error mapping function was subsequently constructed. Curvature values ​​were divided into three clinically relevant intervals: the high curvature interval corresponds to the dangerous segment of the pedicle inflection, with a maximum permissible error of 0.05 mm; the medium curvature interval has a permissible error of 0.10 mm; and the low curvature interval has a permissible error relaxed to 0.20 mm. Linear interpolation was used for smoothing in the interval transition zones to avoid abrupt changes in reconstruction accuracy. This function was integrated into the mesh generation algorithm in the form of a lookup table: when a triangular facet's curvature deviates from the centerline reference value during reconstruction iteration, the optimization weights are dynamically adjusted based on the permissible error threshold of the corresponding curvature interval: a tenfold penalty coefficient is applied to the high curvature interval to force convergence, while computational resource allocation is reduced in the low curvature interval.

[0032] The curvature-allowable error mapping function described above satisfies the following relationship: in, For the maximum permissible geometric error, The discrete curvature of the pedicle centerline. For high curvature threshold, The low curvature threshold.

[0033] S3. Combine the physical lumbar spine model and the three-dimensional lumbar spine model to obtain the virtual-real registration result of the lumbar spine.

[0034] Specifically, S3 includes the following steps: S31. Spatial registration is performed by combining the physical lumbar spine model and the three-dimensional lumbar spine model to obtain the initial registration result of the lumbar spine.

[0035] In this embodiment, multimodal recognition is performed on the cubic marker to obtain near-infrared features of its edges and geometric contour features to construct a composite feature point set; a virtual-real registration state equation is established based on the composite feature point set, and a virtual-real registration matrix is ​​obtained according to the virtual-real registration state equation; a deformation displacement field of the cubic marker is constructed based on the local elastic deformation, and the virtual-real registration matrix is ​​updated based on the deformation displacement field to obtain the initial registration result of the lumbar spine.

[0036] In the initial stage of spatial registration, a multi-sensor collaborative system of an augmented reality head-mounted display captures a cubic marker on the physical model. A camera focuses on the near-infrared reflective coating embedded in the marker's edges, extracting high-contrast light spot features under surgical lighting interference. Simultaneously, a LiDAR sensor emits a structured laser point cloud, scanning the marker surface to generate 3D coordinate points. A random sampling consensus algorithm is used to fit the cube's geometric contour edges. The feature points acquired from the two modalities are fused into a composite feature point set, synchronized with the coordinate transformation matrix using timestamps, forming a composite feature point set that includes spatial location, reflection intensity, and geometric topological relationships.

[0037] Furthermore, a virtual-real registration state equation is constructed based on a composite feature point set. Using the pre-stored vertex coordinates of markers in the 3D model as the target position and the set of entity feature points as the observation position, a least-squares optimization problem is established: the objective function simultaneously constrains the orthogonality of the rotation matrix and the magnitude range of the translation vector, and introduces the gravity direction obtained from the AR device as an additional constraint term. A nonlinear iterative algorithm is used to solve for the optimal rigid body transformation matrix, and the step size is dynamically adjusted after each iteration based on the reprojection error. After solving, a 6-DOF virtual-real registration matrix is ​​output.

[0038] The above-mentioned virtual-real registration state equations satisfy the following relationship: in, This represents the spatial coordinate vector of composite feature points in a three-dimensional lumbar spine skeleton model. Let be a rotation matrix. This refers to the spatial coordinate vector of composite feature points in a physical lumbar spine skeleton model. It is a translation vector. This is the error vector.

[0039] In this embodiment, to overcome registration drift caused by tissue deformation under pressure during surgery, a deformation displacement field of the tissue surrounding the marker is established based on local elasticity. A micro-force sensor array is embedded in the agar layer of the solid model to monitor the pressure distribution data during drilling operations in real time. The displacement field is derived based on Hooke's law: with the marker as the center, the displacement is directly proportional to the pressure value and inversely proportional to the square of the distance, and the elastic modulus of agar is introduced as an attenuation coefficient. When the maximum drilling pressure exceeds 8N, the system automatically calculates the offset vector of the marker in the X / Y / Z axes and generates a 3×3 displacement gradient matrix. The virtual-real registration matrix is ​​updated based on the displacement gradient matrix to obtain the initial registration result of the lumbar vertebrae, and the updated registration matrix compensates for deformation errors. Finally, the virtual pedicle contour and the solid skeleton are projected onto the AR interface to achieve the superposition effect.

[0040] The above deformation displacement field satisfies the following relationship: in, For deformation displacement field, This is the pressure value. The elastic modulus of agar. The cross-sectional area of ​​the contact area between the cubic marker and the solid lumbar spine model. The base of the natural logarithm, The attenuation coefficient is... This represents the distance from the cube marker.

[0041] The updated registration matrix satisfies the following relationship: in, The updated registration matrix, For virtual-real registration matrix, For the deformation displacement field in Components in the axial direction, For the deformation displacement field in Components in the axial direction, For the deformation displacement field in Components in the axial direction.

[0042] S32. Dynamically correct the initial registration result of the lumbar vertebrae to obtain the virtual-real registration result of the lumbar vertebrae.

[0043] In this embodiment, a dynamic correction engine is constructed, and the pose drift vector of the initial registration result of the lumbar spine is obtained based on the dynamic correction engine. A pose feature backtracking equation is constructed based on the pose drift vector, and a correction registration matrix is ​​obtained by combining the initial registration result of the lumbar spine. The virtual-real overlap error is obtained based on the composite feature point set, and the correction registration matrix is ​​optimized by negative feedback based on the virtual-real overlap error to obtain the virtual-real registration result of the lumbar spine.

[0044] Specifically, a dynamic correction engine is constructed with real-time perception and dynamic control at its core. First, a multi-source sensor data acquisition module is integrated. A near-infrared camera continuously captures the angular feature signals of a cubic marker, while an optical tracking device simultaneously records the spatial coordinate changes of composite feature points in both the physical and 3D lumbar spine models, forming a continuous pose data stream. Based on this, a feature matching and deviation analysis unit is built to compare the real-time acquired feature point positions with the reference positions in the initial registration state, calculating the pose drift vector reflecting the spatial deviation between the two. Simultaneously, an adaptive filtering module is embedded to filter noise from the acquired raw data, eliminating outliers caused by equipment jitter, lighting interference, etc., ensuring the stability and accuracy of the drift vector. To achieve dynamic response, the engine incorporates a real-time calculation unit. Based on preset correction logic, the pose drift vector is converted into executable adjustment commands, synchronously linking the 3D model's rendering module and the physical model's positioning system, enabling the virtual model to correct its spatial position in real time based on the drift vector. In addition, a feedback evaluation mechanism is designed to continuously compare the overlap of the corrected feature points and dynamically optimize the filtering parameters and feature matching thresholds to ensure that the engine can maintain efficient deviation identification and correction capabilities under different operating scenarios, ultimately forming a closed-loop dynamic control system to achieve real-time correction of the initial registration results of the lumbar spine.

[0045] The pose drift vectors described above satisfy the following relationship: in, This is the amount of translational drift. The number of feature points, For the index variable of the feature points, For the cube marker The spatial coordinates of each feature point at the current time. For the cube marker The spatial coordinates of each feature point when the initial registration is completed. For the rotational drift angle, Angle calculation function, Let be the rotation matrix at the current moment. This is the rotation matrix when the initial registration is complete.

[0046] Furthermore, a pose feature backtracking equation is constructed based on the pose drift vector, and a correction registration matrix is ​​generated. This is then dynamically extrapolated by combining historical registration data with the current drift state. First, the pose drift vector sequence over a period after initial registration is extracted from the stored logs. A time-series analysis algorithm is used to identify drift patterns, and a pose feature backtracking equation containing a time factor and cumulative drift is constructed. This equation can backtrack and calculate the historical cumulative impact of the deviation based on the current drift vector, thereby determining the adjustment amount that needs to be compensated in reverse. Subsequently, this adjustment amount is combined with the rotation matrix and translation vector from the initial registration result. A preliminary correction registration matrix is ​​generated through spatial transformation operations. The geometric logic of the three-dimensional coordinate transformation ensures that the compensation amounts for rotation and translation can accurately offset the deviation caused by pose drift. For example, when a continuous drift along the positive X-axis is detected in the solid lumbar spine model, the correction matrix will include a translation compensation amount along the negative X-axis. Simultaneously, based on the angle of rotational drift, the Euler angle parameters in the rotation matrix are adjusted accordingly, enabling the three-dimensional lumbar spine model to inversely track the solid lumbar spine model.

[0047] The above pose feature backtracking equation satisfies the following relationship: in, for The state vector at time t, Here is the state transition matrix. for The state vector at time t, For process noise, for The observation vector at time t, For the observation matrix, To observe noise.

[0048] In this embodiment, calculating the virtual-real overlap error based on the composite feature point set and performing negative feedback optimization is a key closed-loop step in improving registration accuracy. First, a dynamic correction engine is used to obtain the current coordinates of composite feature points in the physical and virtual models in real time. Using the spatial distance calculation formula, the Euclidean distance between each pair of corresponding feature points is calculated one by one. The average and maximum values ​​of these distances are used as quantitative indicators to measure the virtual-real overlap error. Using the virtual-real overlap error as the objective function, the rotation and translation parameters in the correction registration matrix are adjusted using a gradient descent algorithm to minimize the error index. The optimized correction registration matrix is ​​then reapplied to the virtual-real registration system. Simultaneously, the system continuously monitors error changes. If the error does not decrease to within the threshold after multiple iterations, the step size and number of iterations of the optimization algorithm are automatically adjusted until the virtual-real overlap error stabilizes within an acceptable range (usually controlled within 0.5 mm), ultimately resulting in a stable and highly accurate virtual-real registration result for the lumbar spine.

[0049] The aforementioned virtual-real overlap error satisfies the following relationship: in, This represents the overlap error value between real and virtual objects. The number of feature points, For the index variable of the feature points, These are the weighting coefficients. For virtual feature point coordinates, These are the coordinates of the entity's feature points.

[0050] S4. Based on the lumbar vertebral bone registration results, the real-time path trajectory of the lumbar pedicle screw is obtained to achieve dynamic guidance of the lumbar pedicle screw.

[0051] In this embodiment, pedicle channel parameters are obtained based on a three-dimensional lumbar spine model. These parameters include the radius of curvature of the pedicle axis and the thickness of the cortical bone. The insertion angle of the lumbar pedicle screw is obtained by combining the virtual and real registration results of the lumbar spine and the pedicle channel parameters. The surgical instrument pose is obtained, and the instrument-planned path deviation angle is obtained based on the surgical instrument pose. The insertion angle is dynamically updated according to the instrument-planned path deviation angle to obtain the real-time path trajectory.

[0052] When obtaining pedicle access parameters based on a 3D lumbar spine model, the reconstructed 3D model is refined using medical image processing software. For the pedicle axis curvature radius, a curve fitting algorithm is used to extract the pedicle central axis, which is discretized into several sampling points. Then, a spatial curve is fitted using the least squares method, and the curvature radius of this curve is calculated. For cortical bone thickness, through layer-by-layer analysis of the 3D model, the vertical distance between the inner and outer surfaces of the cortex is measured at key locations such as the pedicle inlet, middle, and outlet. The average value of multiple measurement points is taken as the final thickness parameter to ensure that subsequent screw insertion angle planning avoids weak areas of the cortex.

[0053] The screw insertion angle is obtained by combining the virtual and real registration results of the lumbar vertebrae and the pedicle channel parameters. First, the coordinate system of the 3D model is aligned with the spatial coordinate system of the lumbar vertebrae in the solid model using the virtual and real registration results, ensuring that the virtual parameters are accurately mapped to the solid space. Based on this, the overall path is determined according to the radius of curvature of the pedicle axis, and the ideal trajectory for screw implantation is designed with the axis as the center. Simultaneously, the initial screw insertion angle is calculated at the insertion point, taking into account the cortical bone thickness. This angle must ensure that the screw axis forms a suitable incident angle with the cortical bone surface, avoiding cortical penetration while allowing it to advance along the center of the pedicle channel. Through spatial geometric calculations, these angle parameters are converted into the operating angles of the surgical instruments, forming the preliminary screw insertion angle, satisfying the following relationship: in, For dynamically optimized nail insertion angle, Based on the basic nail-feeding angle, This is the curvature compensation coefficient. The radius of curvature of the pedicle axis is denoted as .

[0054] Acquiring the surgical instrument pose and calculating the instrument-planned path deviation angle relies on the precise monitoring of a real-time tracking system. The system uses positioning markers attached to the surgical instruments, combined with optical tracking equipment, to capture the instrument's three-dimensional spatial coordinates and attitude information in real time, including the instrument tip position and axial direction. The real-time pose data is compared with the preset planned path, and the deviation angle between the instrument's current axis and the planned path is calculated using a vector angle calculation method. This angle is decomposed into multiple dimensions to clearly reflect the specific direction and degree of instrument deviation. For example, when the instrument tilts inward, the horizontal deviation angle is positive, and when it tilts outward, it is negative, providing a clear quantitative basis for subsequent angle adjustments.

[0055] The above-mentioned instrument-planned path deviation angle satisfies the following relationship: in, For the deviation angle of the planned path of the instrument, It is an inverse cosine function. This is the actual direction vector of the instrument. This is the direction vector for the planned path.

[0056] The insertion angle is dynamically updated based on the instrument-planned path deviation angle, generating a real-time path trajectory. The calculated deviation angle is used as a feedback signal to correct the initial insertion angle using a preset adjustment algorithm: if the deviation angle is within the allowable range (usually less than 1°), the angle parameters are fine-tuned to gradually return the instrument to the planned path; if the deviation is large (greater than 5°), a compensation angle is generated based on the deviation direction to ensure that the adjusted insertion angle can offset the current deviation. The updated angle parameters are transmitted in real-time to the augmented reality guidance interface, displayed as dynamic lines or virtual guide boxes, intuitively guiding the surgeon to adjust the instrument posture. This process requires high-frequency updates (usually more than 30 times per second) to ensure that the path trajectory follows the instrument's pose changes in real time, ultimately forming a real-time guidance path synchronized with instrument operation and adapted to anatomical structures, satisfying the following relationship: in, This is the updated path direction vector. This is the actual direction vector of the instrument. This is the proportionality coefficient. This is the direction vector for the planned path.

[0057] In this embodiment, the minimum distance between the screw axis and the cortical bone is obtained based on the real-time path trajectory; the screw insertion pressure value is obtained, and the speed of the lumbar pedicle screw is adjusted based on the screw insertion pressure value to obtain the screw insertion speed; a real-time early warning mechanism for perforation risk is established based on the minimum distance between the screw axis and the cortical bone and the screw insertion speed; and the lumbar pedicle screw is dynamically guided according to the real-time early warning mechanism for perforation risk.

[0058] Specifically, the screw axis (i.e., the central axis of the real-time path trajectory) is parametrically represented in three-dimensional space. The cortical bone model is meshed, and discrete point clouds are extracted from the surface. The spatial distance from each point to the screw axis is calculated one by one. The minimum value is the minimum distance between the screw axis and the cortical bone, which satisfies the following relationship: in, The minimum distance between the screw axis and the bone cortex. This indicates taking the minimum value. Let be any three-dimensional coordinate point on the surface of the bone cortex. Indicates the surface of the bone cortex. The equation of the screw axis is... For the nail tip coordinates, This is the updated path direction vector.

[0059] It should be noted that the screw axis equation is a straight line in space that starts from the screw tip and extends along the screw insertion direction, which fully describes the axial position and direction of the screw in three-dimensional space.

[0060] In this embodiment, the process of acquiring the screw insertion pressure value and adjusting the screw insertion speed relies on the linkage between the pressure sensor integrated on the surgical instrument and the power control system. The pressure sensor collects the pressure signal generated by the screw contacting bone tissue during implantation in real time, converts it into an electrical signal, and transmits it to the control system. When the pressure value is below a threshold (e.g., 5N), it is determined that the current location is in the cancellous bone region, and the screw insertion speed is appropriately increased to improve efficiency; when the pressure value exceeds the threshold (e.g., 15N), it indicates that the screw may be approaching the cortical bone or encountering hard bone, and the screw insertion speed is immediately reduced or even stopped.

[0061] Furthermore, a real-time early warning mechanism for perforation risk needs to be established based on the minimum distance between the screw axis and the cortical bone and the screw insertion speed. This requires setting multi-dimensional risk assessment indicators. The minimum distance is divided into several levels: distance > 2mm is low risk, distance 1mm-2mm is medium risk, and distance < 1mm is high risk. Simultaneously, combined with the screw insertion speed, when a high-risk level is combined with high-speed screw insertion (e.g., > 5mm / s), a level one warning is triggered (audio-visual alarm + instrument deceleration); when a medium-risk level is combined with medium-speed screw insertion (3mm / s-5mm / s), a level two warning is triggered (visual cues + speed fine-tuning). The early warning mechanism will also dynamically optimize the threshold based on historical data. For example, for areas with thin cortical bone, the high-risk distance threshold will be automatically lowered to improve warning sensitivity.

[0062] The system dynamically guides the screw based on a real-time perforation risk warning mechanism, employing multi-layered intervention measures. When a Level 1 warning is triggered, the danger zone is highlighted on the augmented reality interface, and a deceleration command is simultaneously sent to the surgical instrument's power system until the risk is eliminated. For a Level 2 warning, the interface displays deviation correction suggestions, such as "adjust 0.5° to the left," assisting the surgeon in manually adjusting the screw's insertion direction. Simultaneously, the system continuously tracks screw position changes, updating the minimum distance and insertion speed every 50ms, and switching warning states in real-time based on risk level changes. If the risk continues to escalate after a warning, the system can trigger emergency braking to prevent the screw from perforating the cortical bone, ultimately achieving fully controllable dynamic guidance.

[0063] S5. Construct a verification experiment to verify the effect of the dynamic guidance of the lumbar pedicle screw.

[0064] In this embodiment, a total of 40 lumbar pedicle screws were implanted. After all lumbar pedicle screws were implanted, CT imaging was performed. The implantation accuracy of each lumbar pedicle screw was evaluated based on the imaging results. The results were classified into three grades according to the degree of pedicle penetration: Grade I (no pedicle penetration), Grade II (screw penetration ≤ 2 mm of cortical bone), and Grade III (screw penetration > 2 mm of cortical bone).

[0065] In automatic tracking mode, the overlap accuracy between the 3D model and the solid skeleton model is 1mm ± 0.8mm, and it is unaffected by surgeon movement or surgical instrument obstruction; the automatic correction time after model offset is 3s ± 0.8s, as shown in Table 1: Table 1 The augmented reality-guided software "Surgeon" has a high-precision automatic identification and positioning function. In all augmented reality-guided screw placement surgeries, 18 screw placements were rated as Grade I, 14 as Grade II, and 8 as Grade III; a total of 32 screw placements were acceptable, as shown in Table 2. Table 2 Please see Figure 2 In one optional embodiment, the present invention provides an augmented reality-based dynamic guidance system for lumbar pedicle screws. The system includes an input device, an output device, a processor, and a memory, all interconnected. The memory stores a computer program comprising program instructions, and the processor is configured to invoke the program instructions to execute specific steps as described in the relevant embodiments of the augmented reality-based dynamic guidance method for lumbar pedicle screws provided by the present invention. The augmented reality-based dynamic guidance system for lumbar pedicle screws provided by the present invention has a complete and stable structure, enhancing the overall applicability and practical application capability of the present invention.

[0066] In summary, the present invention provides a dynamic guidance method and system for lumbar pedicle screws based on augmented reality. It utilizes a 3D-printed solid bone model and establishes a spatial coordinate system by fixing a cubic marker on the iliac crest. Scanning data is obtained by dynamically adjusting the scanning angle based on the coordinate system direction, and a lightweight 3D model is reconstructed using pedicle curvature constraints. Virtual-real spatial registration is achieved using multimodal feature recognition and deformation displacement field compensation, and intraoperative registration accuracy is maintained through a dynamic correction engine. The screw insertion angle is dynamically generated based on pedicle channel parameters, and the screw trajectory is updated in real time by combining the instrument pose deviation angle. Simultaneously, a real-time perforation risk warning mechanism enables dynamic screw guidance. The method of this invention is easy to understand, computationally simple, requires minimal workload, and is convenient for engineering applications, providing a theoretical foundation and technical support for the further development of intelligent medical technology.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A dynamic guidance method for lumbar pedicle screws based on augmented reality, characterized in that, Includes the following steps: Construct a physical lumbar spine model of the patient and obtain cubic markers of the physical lumbar spine model to establish a spatial coordinate system for the lumbar spine. Based on the lumbar spine spatial coordinate system, the scanning data of the physical lumbar spine model is obtained, and the three-dimensional lumbar spine model is reconstructed based on the scanning data. The virtual-real registration result of the lumbar spine is obtained by combining the physical lumbar spine model and the three-dimensional lumbar spine model; The real-time path trajectory of the lumbar pedicle screw is obtained based on the virtual-real registration result of the lumbar vertebrae, so as to realize the dynamic guidance of the lumbar pedicle screw.

2. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 1, characterized in that, The process of constructing a physical lumbar spine model of the patient and obtaining cubic markers from the physical lumbar spine model to establish a spatial coordinate system for the lumbar spine includes: The patient's lumbar vertebral bone structure was obtained, and a solid lumbar vertebral bone model was created based on the lumbar vertebral bone structure using 3D printing technology. The cube marker is fixed on the iliac crest of the solid lumbar spine model, and a near-infrared reflective coating is embedded in the corners of the cube marker; The geometric center of the cube marker is used as a spatial positioning reference point to establish the spatial coordinate system of the lumbar spine.

3. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 1, characterized in that, The step of acquiring scan data of the physical lumbar spine model based on the lumbar spine spatial coordinate system and reconstructing a three-dimensional lumbar spine model based on the scan data includes: The scanning angle of the X-ray is dynamically adjusted based on the direction of the lumbar spine spatial coordinate system to obtain the scanning data; The scanned data is segmented to extract the cortical boundary of the pedicle bone of the solid lumbar spine model, and the cortical boundary of the pedicle bone is used as the segmentation result. Obtain curvature constraints, and reconstruct the three-dimensional lumbar spine skeleton model by combining the segmentation results and the curvature constraints.

4. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 3, characterized in that, The obtained curvature constraint conditions include: Based on the segmentation results, the pedicle centerline is extracted, and the discrete curvature of the pedicle centerline is obtained to construct the pedicle curvature field. A curvature-allowable error mapping function is constructed based on the pedicle curvature field, and the curvature-allowable error mapping function is used as the curvature constraint condition.

5. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 1, characterized in that, The process of obtaining the lumbar spine registration result by combining the physical lumbar spine model and the three-dimensional lumbar spine model includes: The initial registration result of the lumbar spine skeleton is obtained by spatial registration of the physical lumbar spine skeleton model and the three-dimensional lumbar spine skeleton model. The initial registration result of the lumbar vertebrae is dynamically corrected to obtain the virtual-real registration result of the lumbar vertebrae.

6. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 5, characterized in that, The initial registration result of the lumbar spine skeleton is obtained by spatial registration of the physical lumbar spine skeleton model and the three-dimensional lumbar spine skeleton model, including: Multimodal recognition is performed on the cubic marker to obtain near-infrared features of its edges and geometric contour features in order to construct a composite feature point set; A virtual-real registration state equation is established based on the composite feature point set, and a virtual-real registration matrix is ​​obtained based on the virtual-real registration state equation. The deformation displacement field of the cubic marker is constructed based on the local elastic deformation, and the virtual-real registration matrix is ​​updated based on the deformation displacement field to obtain the initial registration result of the lumbar spine.

7. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 6, characterized in that, The process of dynamically correcting the initial registration result of the lumbar vertebrae to obtain the virtual-to-real registration result of the lumbar vertebrae includes: A dynamic correction engine is constructed, and the pose drift vector of the initial registration result of the lumbar vertebrae is obtained based on the dynamic correction engine; Based on the pose drift vector, a pose feature backtracking equation is constructed, and a correction registration matrix is ​​obtained by combining the initial registration results of the lumbar vertebrae. Based on the composite feature point set, the virtual-real overlap error is obtained. The virtual-real overlap error is then used to perform negative feedback optimization on the correction registration matrix to obtain the virtual-real registration result of the lumbar spine.

8. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 1, characterized in that, The real-time path trajectory of the lumbar pedicle screw obtained based on the lumbar vertebral bone registration results includes: Based on the three-dimensional lumbar spine model, pedicle channel parameters are obtained, including the radius of curvature of the pedicle axis and the thickness of the cortical bone. The insertion angle of the lumbar pedicle screw is obtained by combining the results of the lumbar vertebral bone registration with the pedicle channel parameters. Obtain the surgical instrument pose, and based on the surgical instrument pose, obtain the instrument-planned path deviation angle; The real-time path trajectory is obtained by dynamically updating the nail insertion angle based on the instrument-planned path deviation angle.

9. The augmented reality-based dynamic guidance method for lumbar pedicle screws according to claim 8, characterized in that, The method of achieving dynamic guidance of the lumbar pedicle screw includes: The minimum distance between the screw axis and the bone cortex is obtained based on the real-time path trajectory. Obtain the insertion pressure value, and adjust the speed of the lumbar pedicle screw based on the insertion pressure value to obtain the insertion speed; A real-time early warning mechanism for perforation risk is established based on the minimum distance between the screw axis and the bone cortex and the screw insertion speed; The lumbar pedicle screws are dynamically guided according to the real-time early warning mechanism for perforation risk.

10. A dynamic guidance system for lumbar pedicle screws based on augmented reality, characterized in that, The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the augmented reality-based dynamic guidance method for lumbar pedicle screws as described in any one of claims 1-9.

Images