A method and device for autonomous cutting of spinal lamina and stopping upon cutting through

Abstract

The present application relates to a kind of spinal lamina autonomous cutting and its cut through and stop perception method and device, wherein the method comprises: obtaining initial lamina structure data set input to lamina bone segmentation network model, output lamina bone tissue segmentation result and probability distribution diagram of dural sac safety termination surface;According to lamina bone tissue segmentation result, rigid registration relationship constructs millimeter level cuttable feasible region;With probability distribution diagram as the insurmountable safety limit of millimeter level cuttable feasible region, generate six degrees of freedom tool center point motion trajectory, obtain cutting path planning data;According to cutting path planning data, execute autonomous cutting operation under impedance control;According to the determination result, trigger cut-through determination flag signal;Bottom motion controller responds to cut-through determination flag signal.The present application generates six degrees of freedom tool center point motion trajectory by setting probability distribution diagram of dural sac safety termination surface as insurmountable safety limit, significantly reduces the risk of surgery.

Description

Technical Field

[0001] This invention relates to the field of spinal surgery assistance technology, specifically to a method and device for autonomous cutting of the spinal lamina and a computer device for sensing and stopping the cutting once it is completed. Background Technology

[0002] Laminectomy is used to relieve nerve compression caused by herniated discs, spinal stenosis, or other lesions. Traditional laminectomy relies on the surgeon's experience and skill, using manual tools to cut the bone tissue of the vertebral lamina. Manual operation carries certain risks, such as accidental damage to the dural sac or surrounding important structures. Therefore, developing a system capable of autonomously cutting the vertebral lamina with a cut-through-stop sensing capability can reduce surgical risks, particularly the possibility of damage to critical anatomical structures near the surgical area (such as the dural sac).

[0003] Currently, several assistive technologies are used in laminectomy in spinal surgery. For example, three-dimensional models are created using computed tomography (CT) or magnetic resonance imaging (MRI) data, and the position of the tracking device guides the surgeon's operation in real time. However, these technologies cannot directly participate in the actual cutting process. Some robotic systems (such as the da Vinci Surgical System) have begun to be used in spinal surgery, but they lack automatic identification and stopping functions to prevent serious complications such as cutting through the dural sac. Ultrasonic cutters and electrocoagulation devices can reduce bleeding and tissue damage to some extent, but they cannot provide sufficient feedback to help determine when to stop cutting, especially when approaching the dural sac.

[0004] To address the aforementioned issues, this invention proposes a method for autonomous cutting of the spinal lamina and a sensing method for stopping the cutting once it is completed. By setting a probability distribution map of the safe termination surface of the dural sac as an insurmountable safety boundary, a six-degree-of-freedom tool center point motion trajectory is generated to reduce surgical risks. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method and device for sensing autonomous cutting of the spinal lamina and stopping the cutting immediately, as well as a computer device.

[0006] According to one aspect of the present invention, a sensing method for autonomous cutting of the spinal lamina and stopping the cutting upon penetration is provided, comprising:

[0007] An optical coherence tomography probe mounted on the end effector of a surgical robot is used to perform real-time three-dimensional volumetric scanning of the target lamina region to obtain an initial lamina structure dataset. The initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone mineral density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel.

[0008] The initial lamina structure dataset is input into the lamina bone segmentation network model, which outputs the lamina bone tissue segmentation results and the probability distribution map of the safe termination surface of the dura mater. Based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images, and intraoperative scan images, a millimeter-level machinable feasible region is constructed. The probability distribution map of the safe termination surface of the dura mater is used as the insurmountable safety boundary of the millimeter-level machinable feasible region. Based on the trajectory planning algorithm, a six-degree-of-freedom tool center point motion trajectory that meets the requirements of robot arm joint limit, obstacle avoidance, and cutting efficiency is generated to obtain cutting path planning data.

[0009] The autonomous cutting operation under impedance control is executed according to the cutting path planning data; wherein, during the cutting process, real-time sensing data streams including real-time cutting force vector, acoustic emission energy characteristics and dynamic tomographic image stream are collected in real time; multi-condition joint logic judgment is performed based on the real-time sensing data stream, and the cutting through judgment flag signal is triggered according to the judgment result;

[0010] In response to the cut-through determination flag signal, the bottom motion controller immediately interrupts the current trajectory interpolation process, stops the feed motion of the cutting tool, and automatically retracts along the local normal direction of the lamina.

[0011] In one alternative approach, obtaining an initial lamina structure dataset by performing a real-time three-dimensional volumetric scan of the target lamina region using an optical coherence tomography probe mounted on the end effector of the surgical robot further includes:

[0012] The lamina-dura mater interface in the surgical field was acquired layer by layer using a frequency domain OCT system, and a high-fidelity data cube was generated by a fast Fourier transform reconstruction algorithm. Each voxel in the data cube carries calibrated bone tissue backscattering reflection intensity information.

[0013] The local optical signal attenuation coefficient is calculated based on the backscattering reflection intensity information of bone tissue in the data cube. Based on the local optical signal attenuation coefficient, the local signal attenuation characteristics and bone density proxy index that are highly correlated with the microstructure and mineralization degree of bone trabecular bone are derived.

[0014] The upper and lower cortical boundaries of the lamina were extracted using a surface curvature-guided horizontal set segmentation method, and the lamina thickness distribution was calculated point by point. The strong reflective layer of the epidural membrane was identified by the high-contrast imaging capability of the frequency domain OCT system at the interface between soft and hard tissues. The Euclidean distance field between the strong reflective layer of the epidural membrane and the inner lamina was determined by the subvoxel edge localization method, and an anatomical gap distance atlas between the lamina and the dural sac was generated. The anatomical gap distance atlas includes anatomical gap distance and anatomical structure labels.

[0015] In one alternative approach, the lamina segmentation network model includes:

[0016] The voxel embedding layer is used to map the initial lamina structure dataset to a high-dimensional latent space through linear projection, generating an initial voxel embedding tensor with anatomical semantic priors.

[0017] The pyramid dense encoder layer includes a DenseNet-121 backbone structure, wherein the DenseNet-121 backbone structure adopts a fourth-order downsampling path, each order containing densely connected convolutional blocks and transition layers; it is used to output multi-scale encoded feature maps from coarse to fine based on the initial voxel embedding tensor.

[0018] The feature decoder layer adopts the nested skip connection structure of U-Net++ to upsample the multi-scale encoded feature map step by step and output the anatomical semantic feature map.

[0019] The dural sac safe termination surface probability prediction head consists of two parallel branches. The first branch is a fully convolutional integrator classifier, which receives the anatomical semantic feature map and outputs the lamina bone tissue segmentation results. The second branch is an uncertainty-aware regression network, which performs multiple forward propagations under the same input to estimate the prediction variance and outputs a probability distribution map of the dural sac safe termination surface. The probability values ​​of the probability distribution map reflect the risk confidence of the cutting tool reaching the front boundary of the dura mater.

[0020] The anatomical constraint regularization layer imposes anatomical prior constraints on the outputs of the two parallel branches during the inference phase. Specifically, the minimum gap distance field between the lamina and dura mater obtained by rigid registration of preoperative CT and intraoperative OCT is used to perform morphological clipping and gradient smoothing on the probability distribution map of the safe termination surface of the dural sac, ensuring that the probability peak is strictly located within the inner lamina.

[0021] In one alternative approach, constructing a millimeter-scale cuttable feasible region based on the lamina bone tissue segmentation results, a rigid registration relationship between preoperative computed tomography images, and intraoperative scan images further includes:

[0022] The preoperative CT image and the intraoperative OCT scan image are spatially aligned using a hybrid rigid registration algorithm, and a six-degree-of-freedom transformation matrix is ​​output.

[0023] The voxel labels in the segmentation results of the vertebral lamina bone tissue are mapped back to the preoperative CT coordinate system through the inverse transformation of the six-degree-of-freedom transformation matrix to generate the preoperative and intraoperative vertebral lamina bone mask.

[0024] Using the preoperative and intraoperative lamina bone mask as a base, the safety margin is gradually reduced point by point along the normal vector field of the inner surface of the lamina to generate a conservative cuttable volume; wherein, the safety margin is determined according to the 95% confidence boundary in the probability distribution map of the safe termination surface of the dural sac.

[0025] Boolean difference operations are performed between the conservative cuttable volume and the key anatomical no-go zones extracted from preoperative CT, including the neural foramen and pedicle, to eliminate potential risk areas and output a millimeter-level cuttable feasible area that simultaneously satisfies the constraints of preserving bone structure integrity and maintaining a safe distance from the dura mater.

[0026] In one alternative approach, performing multi-condition joint logic determination based on the real-time sensed data stream, and triggering a cut-through determination flag signal based on the determination result, further includes:

[0027] Based on real-time sensing data stream, a multi-condition joint logic judgment is performed. When the judgment result simultaneously meets the preset conditions, the cut-through judgment flag signal is triggered.

[0028] One of the preset conditions is: the real-time cutting force vector amplitude decreases by more than a preset threshold and its direction deviates from the bone surface normal.

[0029] The second preset condition is: the acoustic emission energy characteristics show an attenuation inflection point and indicate penetration of the bone-dura mater interface;

[0030] The third preset condition is: the strong reflection layer of the corresponding dural capsule membrane is detected for the first time in the dynamic tomographic image stream and appears continuously for more than three frames at the cutting front, and the distance between it and the current tool position is less than the minimum safe distance of the corresponding position in the dissection gap distance map.

[0031] In one alternative approach, the local signal attenuation characteristics are quantized using an anisotropic attenuation tensor field, which is fused according to the Carter-Hayes bone force model; wherein the expression for the anisotropic attenuation tensor field is:

[0032]

[0033] in, Bone mineral density proxy index derived from OCT; , Each with points The cortical bone neighborhood and the cancellous bone neighborhood are centered. , ; , For the relevant length, , ; , These are the squares of the volumes of the cortical bone neighborhood and the cancellous bone neighborhood, respectively. OCT intensity volume data At point Spatial gradient along the i-th coordinate axis; OCT intensity volume data At point The spatial gradient along the j-th coordinate axis; , For spatial variables; Threshold for bone tissue type discrimination; voxel point The first anisotropic decaying tensor field One component; The coordinates are the spatial positions of the inner plate of the lamina.

[0034] In one alternative approach, the expression for the safety margin is:

[0035]

[0036] in, ; ; This refers to the thickness distribution of the vertebral lamina. The spatial coordinates of the inner plate of the lamina; This is the dividing depth; As the weight core, ; It is a two-dimensional offset vector within the cross section; To correspond to the half-width of the cutting edge of a high-speed drill, ; The confidence coverage threshold. ; Let be a transverse cross-sectional element perpendicular to the normal at depth s; The probability distribution of the safe termination surface of the dural sac is the value of the corresponding spatial location; These are candidate security depth variables.

[0037] In one alternative approach, the penetration confidence of the multi-condition joint logic decision is:

[0038]

[0039] in, For reference cutting force, ; Bone mineral density proxy index derived from OCT; The acoustic emission coefficient; This represents the minimum safe gap in the local area derived from the anatomical gap distance map; It is a Gaussian trust function; This is the real-time cutting force vector; The second time derivative of the acoustic emission energy; For indicator functions; The Euclidean distance from the current tool center point TCP to the strong reflective layer of the dural capsule; The number of consecutive detection frames; For time.

[0040] According to another aspect of the present invention, a sensing device for autonomous cutting of the spinal lamina and stopping upon cutting through is provided, comprising:

[0041] The intraoperative three-dimensional structure perception module is used to perform real-time three-dimensional volume scanning of the target lamina region using an optical coherence tomography probe installed on the end effector of the surgical robot to obtain an initial lamina structure dataset. The initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone mineral density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel.

[0042] The safe cutting region planning module is used to input the initial lamina structure dataset into the lamina bone segmentation network model, and output the lamina bone tissue segmentation results and the probability distribution map of the safe termination surface of the dura mater; based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images and intraoperative scan images, a millimeter-level cuttable feasible region is constructed; the probability distribution map of the safe termination surface of the dura mater is used as the insurmountable safety boundary of the millimeter-level cuttable feasible region, and a six-degree-of-freedom tool center point motion trajectory that meets the requirements of robot arm joint limit, obstacle avoidance and cutting efficiency is generated according to the trajectory planning algorithm to obtain the cutting path planning data;

[0043] The cutting state perception and penetration discrimination module is used to perform autonomous cutting operation under impedance control according to the cutting path planning data; wherein, during the cutting process, real-time perception data stream including real-time cutting force vector, acoustic emission energy characteristics and dynamic tomographic image stream is collected in real time; multi-condition joint logic judgment is performed based on the real-time perception data stream, and a penetration judgment flag signal is triggered according to the judgment result;

[0044] The retraction execution module is used by the underlying motion controller to immediately interrupt the current trajectory interpolation process, stop the feed motion of the cutting tool, and automatically retract along the local normal direction of the vertebral plate in response to the cut-through determination flag signal.

[0045] According to another aspect of the present invention, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0046] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-mentioned autonomous spinal lamina cutting and its sensing method of stopping after cutting through.

[0047] According to the solution provided by the present invention, a real-time three-dimensional volumetric scan of the target lamina region is performed using an optical coherence tomography (OCT) probe mounted on the end effector of a surgical robot to obtain an initial lamina structure dataset. This initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel. The initial lamina structure dataset is input into a lamina bone segmentation network model, which outputs the lamina bone tissue segmentation results and a probability distribution map of the safe termination surface of the dural sac. A millimeter-level cuttable feasible region is constructed based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images, and intraoperative scan images. The safe termination surface of the dural sac is then used as the final step. The probability distribution map serves as an insurmountable safety boundary for the millimeter-level cutable feasible area. A six-degree-of-freedom tool center point motion trajectory, satisfying the requirements of robotic arm joint limits, obstacle avoidance, and cutting efficiency, is generated based on the trajectory planning algorithm, resulting in cutting path planning data. Autonomous cutting operations under impedance control are executed based on this cutting path planning data. During the cutting process, real-time sensing data streams, including real-time cutting force vectors, acoustic emission energy characteristics, and dynamic tomographic image streams, are acquired. Multi-condition joint logic judgments are performed based on these real-time sensing data streams, triggering a cut-through judgment flag signal based on the judgment result. The underlying motion controller, responding to the cut-through judgment flag signal, immediately interrupts the current trajectory interpolation process, stops the cutting tool's feed motion, and automatically retracts along the local normal direction of the vertebral lamina. This invention significantly reduces surgical risks by setting a probability distribution map of the dural sac's safe termination surface as an insurmountable safety boundary and generating a six-degree-of-freedom tool center point motion trajectory. Specifically, optical coherence tomography (OCT) and a vertebral lamina bone segmentation network model are used to accurately identify the boundary between the vertebral lamina bone tissue and the dural sac, improving the safety and accuracy of the surgery. During the cutting process, real-time acquisition of data including cutting force vectors, acoustic emission energy characteristics, and dynamic tomographic image streams is performed, along with multi-condition joint logic judgments to avoid damage to the dural sac. A six-DOF tool center point motion trajectory is generated based on a trajectory planning algorithm, satisfying the robotic arm's joint constraints, obstacle avoidance requirements, and cutting efficiency, significantly reducing surgical risks.

[0048] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0049] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0050] Figure 1 A flowchart illustrating the autonomous cutting of the spinal lamina and its sensing method for stopping the cutting upon completion, according to an embodiment of the present invention, is shown.

[0051] Figure 2 A schematic diagram of a vertebral lamina segmentation network model according to an embodiment of the present invention is shown;

[0052] Figure 3 A schematic flowchart of spinal laminar lamina ablation control and response according to an embodiment of the present invention is shown;

[0053] Figure 4 A schematic diagram of the frame of the sensing device for autonomous cutting of the spinal lamina and its stopping upon cutting through is shown in an embodiment of the present invention.

[0054] Figure 5 A schematic diagram of the structure of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0055] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0056] Figure 1 A flowchart illustrating the autonomous spinal lamina ablation and its sensing method for stopping the ablation upon completion, according to an embodiment of the present invention, is shown. Specifically, as... Figure 1 As shown, it includes the following steps:

[0057] Step S101: The target lamina region is scanned in real time using an optical coherence tomography probe installed on the end effector of the surgical robot to obtain an initial lamina structure dataset; wherein, the initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel.

[0058] In this embodiment, an optical coherence tomography (OCT) probe is mounted on the end effector of the surgical robot, providing an axial resolution better than 10 μm. This allows for clear differentiation of the interfaces between the cortical bone, cancellous bone, and epidural membrane of the vertebral sac, particularly highlighting the strong reflective signal characteristics of critical safety structures within the dural sac, thus providing a reliable basis for setting safe cutting boundaries. For example, a miniature frequency-domain OCT probe is rigidly fixed to the end effector of the spinal surgery robot, maintaining a known spatial relative pose with the cutting tool (such as a high-speed drill). Before surgery begins, the robot guides the OCT probe close to the target vertebral sac region, performing a rapid three-dimensional scan of the surgical field without contact. The acquired initial vertebral sac structure dataset not only contains geometric information (such as vertebral sac thickness distribution and anatomical interstitial distance) but also incorporates physical quantities reflecting the microstructure and mineralization state of bone tissue (such as reflection intensity, signal attenuation characteristics, and bone mineral density surrogate indices). Because the OCT probe is directly installed at the end of the robot, it can continuously scan the surgical field before, during and after the cutting, realizing a closed-loop operation mode of seeing and cutting simultaneously, avoiding the problem of preoperative image failure caused by intraoperative tissue displacement, bleeding or deformation.

[0059] In one alternative approach, obtaining an initial lamina structure dataset by performing a real-time three-dimensional volumetric scan of the target lamina region using an optical coherence tomography probe mounted on the end effector of the surgical robot further includes:

[0060] The lamina-dura mater interface in the surgical field was acquired layer by layer using a frequency domain OCT system, and a high-fidelity data cube was generated by a fast Fourier transform reconstruction algorithm. Each voxel in the data cube carries calibrated bone tissue backscattering reflection intensity information.

[0061] The local optical signal attenuation coefficient is calculated based on the backscattering reflection intensity information of bone tissue in the data cube. Based on the local optical signal attenuation coefficient, the local signal attenuation characteristics and bone density proxy index that are highly correlated with the microstructure and mineralization degree of bone trabecular bone are derived.

[0062] The upper and lower cortical boundaries of the lamina were extracted using a surface curvature-guided horizontal set segmentation method, and the lamina thickness distribution was calculated point by point. The strong reflective layer of the epidural membrane was identified by the high-contrast imaging capability of the frequency domain OCT system at the interface between soft and hard tissues. The Euclidean distance field between the strong reflective layer of the epidural membrane and the inner lamina was determined by the subvoxel edge localization method, and an anatomical gap distance atlas between the lamina and the dural sac was generated. The anatomical gap distance atlas includes anatomical gap distance and anatomical structure labels.

[0063] In this embodiment, layer-by-layer B-scan acquisition generates a data cube containing backscattered reflection intensity information of bone tissue at each voxel point, providing micron-level anatomical details. The local optical signal attenuation coefficient is calculated based on the bone tissue backscattered reflection intensity information, and further, local signal attenuation characteristics and bone mineralization surrogate indices reflecting the microstructure and mineralization degree of bone trabeculae are derived. A surface curvature-guided level set segmentation method is used to not only accurately extract the upper and lower cortical boundaries of the lamina but also to calculate the lamina thickness distribution point-by-point, improving the accuracy of surgical planning. Utilizing the high-contrast imaging capability of the OCT system at the interface between soft and hard tissues, the strongly reflective layer of the epidural membrane can be clearly identified. The Euclidean distance field between this layer and the inner lamina is measured using a subvoxel edge localization method, thereby generating an anatomical gap distance map between the lamina and the dural sac. This ensures that important nerve tissues are not accidentally damaged during surgery, significantly improving surgical safety. For example, in spinal surgery, when surgeons need to remove a portion of the L4 lamina to relieve symptoms caused by lumbar spinal stenosis, the surgical robot uses an OCT probe to perform a detailed scan of the L4 lamina, generating a data cube containing rich information. Analyzing the data cube not only accurately depicts the specific morphology and thickness variations of the lamina, but also allows for the assessment of local bone density and the prediction of potential safety risks. For example, if a lamina is found to be only 1.5 mm thick at a specific location and very close to the dural sac, this area is marked as high-risk, and additional safety margins are recommended when developing the cutting plan. Because detailed anatomical distance maps are available, surgeons can operate with greater confidence during surgery, knowing when to stop cutting to avoid damaging the dural sac or other important structures, thus improving surgical success rates and significantly reducing the risk of complications.

[0064] In one alternative approach, the local signal attenuation characteristics are quantized using an anisotropic attenuation tensor field, which is fused according to the Carter-Hayes bone force model; wherein the expression for the anisotropic attenuation tensor field is:

[0065]

[0066] in, Bone mineral density proxy index derived from OCT; , Each with points The cortical bone neighborhood and the cancellous bone neighborhood are centered. , ; , For the relevant length, , ; , These are the squares of the volumes of the cortical bone neighborhood and the cancellous bone neighborhood, respectively. OCT intensity volume data At point Spatial gradient along the i-th coordinate axis; OCT intensity volume data At point The spatial gradient along the j-th coordinate axis; , For spatial variables; Threshold for bone tissue type discrimination; voxel point The first anisotropic decaying tensor field One component; The coordinates are the spatial positions of the inner plate of the lamina.

[0067] In this embodiment, a bone mineral density surrogate index retrieved from OCT is used as a discrimination threshold to dynamically select neighborhood parameters suitable for cortical or cancellous bone. This allows for appropriate attenuation in different bone tissue types, significantly improving tissue classification accuracy. The constructed second-order tensor field can characterize the differences in signal attenuation rates in different spatial directions, effectively representing the orientation and porosity of trabecular bone and more realistically reflecting the local force intensity distribution of the vertebral lamina. The Carter-Hayes model establishes a power-law relationship between bone mineral density and elastic modulus. By integrating this model into OCT signal processing, the anisotropic attenuation tensor not only has imaging significance but can also be directly used to predict cutting resistance, optimize feed rate, or assess structural stability. By defining it point-by-point in space, this tensor information can be dynamically invoked during trajectory planning and impedance control stages, allowing for more conservative cutting parameters to be applied to regions with high attenuation anisotropy (i.e., weak forces), thus balancing efficiency and safety. Furthermore, enhancing the perception of complex bone microstructures, especially in areas of osteoporosis or pathological variations, helps improve the reliability of predicting the safe termination surface of the dural sac.

[0068] For example, a 70-year-old female patient with osteoporosis underwent L3 laminectomy. Intraoperative OCT scan showed localized areas of the L3 laminectomy. = 0.58 < This area was identified as a predominantly cancellous bone region. Automatic activation is then performed in this region. Calculated The cutting tool exhibits significant anisotropy, with attenuation in one direction being significantly weaker than in others, indicating sparse trabecular bone arrangement along that direction. Based on the Carter-Hayes model, the elastic modulus of this region is estimated to be only about 40% of that of normal cortical bone, predicting a high risk of fragmentation or overcutting during cutting. Therefore, when generating the cutting path, the robot automatically reduces the feed rate in this region and sets a lower upper limit for the reference force in the impedance control. During actual cutting, when the tool approaches this weak area, abnormal fluctuations in cutting force are detected in advance. Combining acoustic emission and OCT dynamic images, the tool triggers penetration and stops immediately 0.2 mm before penetration, avoiding dural damage.

[0069] Step S102: Input the initial lamina structure dataset into the lamina bone segmentation network model, and output the lamina bone tissue segmentation results and the probability distribution map of the safe termination surface of the dura mater; construct a millimeter-level cuttable feasible region based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images and intraoperative scan images; use the probability distribution map of the safe termination surface of the dura mater as the insurmountable safety boundary of the millimeter-level cuttable feasible region, and generate a six-degree-of-freedom tool center point motion trajectory that meets the requirements of robot arm joint limit, obstacle avoidance and cutting efficiency according to the trajectory planning algorithm, and obtain cutting path planning data.

[0070] In this embodiment, for example, the L3-L4 lamina were removed intraoperatively (goal: to relieve spinal cord compression and preserve intervertebral disc stability). After scanning the lamina with an OCT system, the gradient distribution of lamina thickness from 3.8 mm to 5.2 mm was identified, and the strongly reflective layer of the epidural membrane (at a depth of 2.1 mm) was located in the OCT B-scan. Bone tissue segmentation results showed that the safe area with an inner lamina thickness ≥ 2.5 mm accounted for 76% of the area; the probability distribution map of the safe termination surface of the dura mater focused on the inner surface of the lamina (peak probability 0.95), and the distance to the adjacent spinal nerve gap was marked as 2.8 mm by the anatomical gap atlas. After rigid registration, a millimeter-level cuttable region was constructed. After deducting the restricted area within a safety margin of 1.0 mm, the remaining cuttable volume was 185 mm³ (original volume 312 mm³). The trajectory planning algorithm generates a trajectory with the starting point at the outer edge of the vertebral lamina (Z=12.3mm) and the ending point at the safety margin boundary (Z=9.7mm). Constraints include avoiding the right neural foramen (offset ≥1.5mm); limiting the robotic arm joint angle to scapular eversion ≤150°; and maintaining a cutting rate of 0.8mm / s (adapting to the OCT scan frame rate). The generated six-degree-of-freedom path is a helical embedded trajectory with a tool movement trajectory length of 142mm, a 23% reduction compared to traditional straight paths. During cutting, real-time sensing triggers three conditions: the cutting force vector amplitude decreases by 1.2N / s (threshold 0.8N / s); the acoustic emission energy characteristic decay rate is -2.7dB / s (threshold -2.5dB / s); and the dura mater's strong reflective layer appears consecutively in the first three frames of the OCT image at a safety clearance of 0.6mm. The cut-through marker signal is triggered (confidence level 0.97), and the motion controller cuts off the blade feed (instantaneous stop); the blade retracts 0.4 mm along the normal direction (completed within 0.3 seconds); the current OCT frame is saved for postoperative path optimization. The cutting depth is controlled within 2.1 mm (within the safety margin boundary), without touching the dura mater, and the intraoperative blood loss is <5 mL (compared to an average of 18 mL in traditional surgery). This achieves a shift from empirical conservative cutting to precise and controllable millimeter-level autonomous cutting, reducing the risk of intraoperative dura mater penetration from 12.7% to 0.3% in clinical practice, while improving cutting efficiency by 48%.

[0071] In one alternative approach, the lamina segmentation network model includes:

[0072] The voxel embedding layer is used to map the initial lamina structure dataset to a high-dimensional latent space through linear projection, generating an initial voxel embedding tensor with anatomical semantic priors.

[0073] The pyramid dense encoder layer includes a DenseNet-121 backbone structure, wherein the DenseNet-121 backbone structure adopts a fourth-order downsampling path, each order containing densely connected convolutional blocks and transition layers; it is used to output multi-scale encoded feature maps from coarse to fine based on the initial voxel embedding tensor.

[0074] The feature decoder layer adopts the nested skip connection structure of U-Net++ to upsample the multi-scale encoded feature map step by step and output the anatomical semantic feature map.

[0075] The dural sac safe termination surface probability prediction head consists of two parallel branches. The first branch is a fully convolutional integrator classifier, which receives the anatomical semantic feature map and outputs the lamina bone tissue segmentation results. The second branch is an uncertainty-aware regression network, which performs multiple forward propagations under the same input to estimate the prediction variance and outputs a probability distribution map of the dural sac safe termination surface. The probability values ​​of the probability distribution map reflect the risk confidence of the cutting tool reaching the front boundary of the dura mater.

[0076] The anatomical constraint regularization layer imposes anatomical prior constraints on the outputs of the two parallel branches during the inference phase. Specifically, the minimum gap distance field between the lamina and dura mater obtained by rigid registration of preoperative CT and intraoperative OCT is used to perform morphological clipping and gradient smoothing on the probability distribution map of the safe termination surface of the dural sac, ensuring that the probability peak is strictly located within the inner lamina.

[0077] In this embodiment, as Figure 2 As shown, the pyramidal dense encoder layer (DenseNet-121) captures multi-scale features from macroscopic anatomical contours to microscopic trabecular bone structures. Through nested skip connections in U-Net++, it achieves deep fusion of high and low-level features during decoding, ensuring the spatial continuity of segmentation boundaries (especially the bone-dura mater interface) and overcoming the ambiguity problem of traditional segmentation methods at complex thin-layer structures. The probability distribution map of the safe termination surface of the dura mater directly reflects the risk confidence of the tool approaching the dura mater, providing a continuous safety boundary for path planning. For areas with poor image quality or large anatomical variations, a lower confidence level (higher uncertainty) can be output, thus adopting a more conservative strategy and enhancing safety when facing various complex intraoperative situations. The anatomical constraint regularization layer plays a role in the inference stage. It uses the precise minimum gap distance field between the lamina and dura mater obtained by preoperative CT and intraoperative OCT registration to perform morphological clipping and gradient smoothing on the probability map of the original output of the network. This forces the correction of erroneous predictions that the neural network may produce that violate the real anatomical structure (e.g., incorrectly predicting the probability peak to the dura mater outside the lamina), ensuring that the final safety boundary (probability peak) is strictly located inside the lamina bone, eliminating the possibility of planning out of bounds.

[0078] In one alternative approach, constructing a millimeter-scale cuttable feasible region based on the lamina bone tissue segmentation results, a rigid registration relationship between preoperative computed tomography images, and intraoperative scan images further includes:

[0079] The preoperative CT image and the intraoperative OCT scan image are spatially aligned using a hybrid rigid registration algorithm, and a six-degree-of-freedom transformation matrix is ​​output.

[0080] The voxel labels in the segmentation results of the vertebral lamina bone tissue are mapped back to the preoperative CT coordinate system through the inverse transformation of the six-degree-of-freedom transformation matrix to generate the preoperative and intraoperative vertebral lamina bone mask.

[0081] Using the preoperative and intraoperative lamina bone mask as a base, the safety margin is gradually reduced point by point along the normal vector field of the inner surface of the lamina to generate a conservative cuttable volume; wherein, the safety margin is determined according to the 95% confidence boundary in the probability distribution map of the safe termination surface of the dural sac.

[0082] Boolean difference operations are performed between the conservative cuttable volume and the key anatomical no-go zones extracted from preoperative CT, including the neural foramen and pedicle, to eliminate potential risk areas and output a millimeter-level cuttable feasible area that simultaneously satisfies the constraints of preserving bone structure integrity and maintaining a safe distance from the dura mater.

[0083] In this embodiment, a hybrid rigid registration algorithm is used to perform sub-millimeter-level spatial alignment between preoperative CT images, which macroscopically reflect the overall anatomical structure, and intraoperative OCT images, which microscopically reflect real-time surface details. This achieves the fusion of the preoperative global planning map and the intraoperative real-time detail map. The preoperative and intraoperative lamina bone mask is not simply a preoperative CT bone model, but rather maps more precise bone surface information segmented in real-time by intraoperative OCT back to the preoperative CT coordinate system. This ensures that the bone structure basis upon which the planning relies includes both complete preoperative three-dimensional information and incorporates actual morphological changes caused by factors such as patient positioning during surgery, guaranteeing the real-time nature and accuracy of the planning base. The safety margin is not based on fixed empirical values ​​(such as "retain 1.0 mm"), but is dynamically calculated point-by-point based on the 95% confidence boundary in the probability distribution map of the dural sac safe termination surface. In areas where the model predicts high risk (steep probability gradient), the safety margin automatically increases, resulting in a more conservative planning; in areas with clearly lower risk, the safety margin can be appropriately reduced to retain more healthy bone. The safety margin considers the influence range of the cutting edge, optimizing the safety boundary for the specific physical characteristics of the cutting tool, rather than a simple geometric equidistant surface. It not only considers the most important dural sac safety (achieved through the safety margin) but also actively eliminates other critical anatomical no-go zones (such as neural foramina and pedicles) through Boolean difference operations. This multi-layered safety constraint ensures that the generated machinable region avoids damage to the dura mater while also mitigating the risk of injury to important nerves and blood vessels, making the planning results more clinically applicable. The millimeter-level machinable region is a three-dimensional volume model (such as an STL mesh or voxel mask) indicating the spatial range that the robotic arm's tool can cut, transforming the motion planning problem into finding the optimal path within this safe volume. For example, a patient undergoing L4 laminectomy has a thicker and more uniform left laminectomy region and a thinner right laminectomy region close to a large nerve root axis. Traditionally, surgeons might manually draw a rectangular area to be removed on a three-dimensional model based on preoperative CT scans and set a uniform safety boundary (e.g., 2mm) for the entire area. The inability to distinguish between the different bone thicknesses and risks on the left and right sides could lead to insufficient safety margins in thin areas on the right side, or unnecessary bone removal in thick areas on the left side. In this application, intraoperative OCT scans revealed that the patient's position resulted in a narrower gap between the inner surface of the right lamina and the dura mater than shown on CT. In the thick area on the left, the probability distribution map showed a gentle risk gradient, and the calculated safety margin was 1.2 mm. In the thin and narrow area on the right, the risk on the probability distribution map increased rapidly, and the calculated safety margin automatically increased to 2.0 mm (even close to the thickness of the lamina itself) to ensure safety. The adjacent neural foramen on the right side was automatically identified and removed from the conservatively cuttable volume. The resulting cuttable feasible area was a relatively "thick" cutting area on the left side, while on the right side it was very narrow, and there might even be no feasible area due to an excessively large safety margin and exclusion of restricted areas.The guided planning algorithm safely performs autonomous cutting on the left side, while marking the right side as high-risk and suggesting manual fine-tuning or abandoning cutting in that area, thus proactively preventing nerve damage. It transforms vague clinical experience ("leaving enough safety margins") into precise operations, providing a safe (multiple safeguards) and efficient (preserving the operable area) 3D blueprint for control command generation.

[0084] In one alternative approach, the expression for the safety margin is:

[0085]

[0086] in, ; ; This refers to the thickness distribution of the vertebral lamina. The spatial coordinates of the inner plate of the lamina; This is the dividing depth; As the weight core, ; It is a two-dimensional offset vector within the cross section; To correspond to the half-width of the cutting edge of a high-speed drill, ; The confidence coverage threshold. ; Let be a transverse cross-sectional element perpendicular to the normal at depth s; The probability distribution of the safe termination surface of the dural sac is the value of the corresponding spatial location; These are candidate security depth variables.

[0087] In this embodiment, a Gaussian weighted kernel, with the half-width of the high-speed drill bit as the scale, is used to incorporate the actual working range of the cutting tool into the safety boundary, making the safety margin more closely resemble real physical cutting behavior. It clearly distinguishes between cortical bone (dense) and cancellous bone (porous) regions (by depth). Using a boundary condition, each component contributes to the risk of dural sac penetration. A confidence coverage threshold is set to ensure that the selected safety depth safety margin covers at least 95% of the cumulative probability of dural sac risk within the tool's effective area, achieving a high-confidence non-penetration target. The safety margin is strictly limited to a reasonable range, avoiding both excessive conservatism leading to ineffective cutting and jeopardizing the dura mater due to excessively thin lamina.

[0088] Step S103: Perform an autonomous cutting operation under impedance control according to the cutting path planning data; wherein, during the cutting process, real-time sensing data stream including real-time cutting force vector, acoustic emission energy characteristics and dynamic tomographic image stream is collected in real time; multi-condition joint logic judgment is performed based on the real-time sensing data stream, and a cut-through judgment flag signal is triggered according to the judgment result.

[0089] In this embodiment, as Figure 3As shown, through impedance control, the robot's end effector can adjust its movement in real time based on the contact force (pausing slightly or fine-tuning its posture when encountering hard bone), simulating the feel of a skilled surgeon to achieve smooth and precise bone tissue removal. Real-time sensing data is used to continuously verify the execution process of the pre-planned procedure. If a deviation from the expected outcome is detected (such as nearing or already complete penetration), a penetration determination signal is immediately generated, ensuring absolute surgical safety. For example, the robot is grinding a lamina of uneven thickness, with the current cutting point located in an area where thicker cortical bone abruptly transitions to an extremely thin bone plate. If only force sensors are available, the cutting force will drop sharply when the drill suddenly penetrates the extremely thin bone plate into the underlying soft tissue space. However, if the drill encounters ligaments or fatty tissue at the moment of penetration, it may generate instantaneous, directionally chaotic force signals, which can be confused with noise or vibration, causing the system to hesitate or delay judgment. If only acoustic emission sensors are available, the acoustic emission signal will exhibit characteristic changes when the bone plate is penetrated. If the operating room environment is noisy (e.g., due to suction devices, other equipment) or if a small amount of bleeding during penetration alters the sound propagation medium, the signal characteristics may be submerged or distorted. In this application, upon receiving the third satisfied signal, the determiner confirms that all three conditions have been met within the allowable time tolerance. At this point, it is certain that a penetration event has occurred, and the penetration determination flag signal is immediately triggered.

[0090] In one alternative approach, performing multi-condition joint logic determination based on the real-time sensed data stream, and triggering a cut-through determination flag signal based on the determination result, further includes:

[0091] Based on real-time sensing data stream, a multi-condition joint logic judgment is performed. When the judgment result simultaneously meets the preset conditions, the cut-through judgment flag signal is triggered.

[0092] One of the preset conditions is: the real-time cutting force vector amplitude decreases by more than a preset threshold and its direction deviates from the bone surface normal.

[0093] The second preset condition is: the acoustic emission energy characteristics show an attenuation inflection point and indicate penetration of the bone-dura mater interface;

[0094] The third preset condition is: the strong reflection layer of the corresponding dural capsule membrane is detected for the first time in the dynamic tomographic image stream and appears continuously for more than three frames at the cutting front, and the distance between it and the current tool position is less than the minimum safe distance of the corresponding position in the dissection gap distance map.

[0095] In this embodiment, the force vector condition requires the amplitude decrease to exceed a threshold and the direction to deviate from the normal, distinguishing between normal cutting force fluctuations (such as encountering trabecular bone gaps) and true interface penetration (disappearance of resistance and possible slippage). The acoustic emission condition requires the attenuation inflection point and indicates interface penetration. By analyzing the derivative of energy changes or specific frequency band characteristics, the moment when the cutting of dense bone stops is captured, avoiding misjudging the sound of the tool idling or cutting cancellous bone. The visual condition requires that the sound appear in more than three consecutive frames and the distance is less than the minimum safe distance, eliminating interference from image noise, transient artifacts, or non-target reflective layers, ensuring that the observed dura mater is stable and in a dangerous position. For example, in a cervical laminectomy, the robot is grinding the C5 lamina. In the initial stage, the cutting force is stable at 2.5N, the acoustic emission energy rises steadily, and the OCT image only shows bone tissue scattering without a strong reflective layer. When approaching the inner lamina, the OCT detects a suspected bright line in the first frame, but it is not continuous; the force signal is still normal. The bright line reappears in the second frame, but the force does not change significantly, and the acoustic emission has no inflection point and is not triggered. The bright line appears for the third consecutive time in the third frame; simultaneously, the cutting force drops sharply from 2.4N to 1.1N (a decrease of >50%) and deviates by 15°; at the same time, the acoustic emission energy curve shows a clear negative inflection point of the second derivative. At this point, confirming that all three conditions are met, the penetration signal is immediately triggered. The underlying controller stops the feed within milliseconds and drives the tool to retract 0.5mm along the lamina normal. The tool stops precisely at the safety boundary without contacting the dura mater, achieving true penetration and immediate stop with zero damage.

[0096] In one alternative approach, the penetration confidence of the multi-condition joint logic decision is:

[0097]

[0098] in, For reference cutting force, ; Bone mineral density proxy index derived from OCT; The acoustic emission coefficient; This represents the minimum safe gap in the local area derived from the anatomical gap distance map; It is a Gaussian trust function; This is the real-time cutting force vector; The second time derivative of the acoustic emission energy; For indicator functions; The Euclidean distance from the current tool center point TCP to the strong reflective layer of the dural capsule; The number of consecutive detection frames; For time.

[0099] In this embodiment, the reference cutting force Bone mineral density proxy index dynamically dependent on intraoperative OCT inversion This avoids misjudgments in porous or dense bone when using a fixed threshold, significantly improving generalization ability. Simultaneously considering both spatial distance (whether it enters a danger zone) and temporal continuity (whether it occurs stably), transient artifacts or local scattering interference are suppressed. A Gaussian trust function is employed. It is not sensitive to small fluctuations and only outputs high confidence when the force is significantly lower than expected.

[0100] In step S104, the bottom motion controller responds to the cut-through determination flag signal by immediately interrupting the current trajectory interpolation process, stopping the feed motion of the cutting tool, and automatically retracting along the local normal direction of the vertebral plate.

[0101] In this embodiment, once the penetration determination signal is triggered, the underlying motion controller immediately interrupts trajectory interpolation (rather than waiting for the current cycle to end), achieving sub-millisecond-level shutdown. This minimizes the potential impact of the blade on vulnerable neural structures such as the dural sac, significantly improving surgical safety. After stopping feed, the robot does not simply hover but automatically retracts along the local normal direction of the lamina. This direction is perpendicular to the bone surface, allowing for the shortest path away from the danger zone and avoiding scraping or compression of adjacent nerve roots or blood vessels due to lateral movement. This preserves the removed bone while preventing cerebrospinal fluid leakage or nerve damage after penetration, demonstrating good clinical interpretability and acceptability, and balancing automation with human-machine collaboration flexibility. For example, in an L4 laminectomy for a patient with lumbar spinal stenosis, the robot is grinding the inner lamina at a speed of 0.2 mm / s. When the blade just penetrates the cortical bone and approaches the dural sac, a sudden drop in force, an acoustic emission inflection point, and the continuous appearance of a strong OCT reflection layer are detected, triggering the penetration determination signal. Within 2ms, the underlying controller interrupted the planned 0.3mm downward cutting trajectory; stopped the Z-axis feed motor; and simultaneously drove the robotic arm to retract 0.8mm along the current local normal direction of the vertebral lamina (obtained through real-time surface fitting). The blade tip rapidly retreated from a position 0.1mm from the dural sac to a safe zone 0.7mm away, without any contact. This transforms the traditional passive protection relying on the surgeon's reaction into an active defense system for the robotic system, making it particularly suitable for high-risk scenarios with limited field of vision, anatomical variations, or extremely thin bone. It is a key execution guarantee for achieving truly safe and reliable autonomous spinal surgery.

[0102] According to the solution provided by the present invention, a real-time three-dimensional volumetric scan of the target lamina region is performed using an optical coherence tomography (OCT) probe mounted on the end effector of a surgical robot to obtain an initial lamina structure dataset. This initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel. The initial lamina structure dataset is input into a lamina bone segmentation network model, which outputs the lamina bone tissue segmentation results and a probability distribution map of the safe termination surface of the dural sac. A millimeter-level cuttable feasible region is constructed based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images, and intraoperative scan images. The safe termination surface of the dural sac is then used as the final step. The probability distribution map serves as an insurmountable safety boundary for the millimeter-level cutable feasible area. A six-degree-of-freedom tool center point motion trajectory, satisfying the requirements of robotic arm joint limits, obstacle avoidance, and cutting efficiency, is generated based on the trajectory planning algorithm, resulting in cutting path planning data. Autonomous cutting operations under impedance control are executed based on this cutting path planning data. During the cutting process, real-time sensing data streams, including real-time cutting force vectors, acoustic emission energy characteristics, and dynamic tomographic image streams, are acquired. Multi-condition joint logic judgments are performed based on these real-time sensing data streams, triggering a cut-through judgment flag signal based on the judgment result. The underlying motion controller, responding to the cut-through judgment flag signal, immediately interrupts the current trajectory interpolation process, stops the cutting tool's feed motion, and automatically retracts along the local normal direction of the vertebral lamina. This invention significantly reduces surgical risks by setting a probability distribution map of the dural sac's safe termination surface as an insurmountable safety boundary and generating a six-degree-of-freedom tool center point motion trajectory. Specifically, optical coherence tomography (OCT) and a vertebral lamina bone segmentation network model are used to accurately identify the boundary between the vertebral lamina bone tissue and the dural sac, improving the safety and accuracy of the surgery. During the cutting process, real-time acquisition of data including cutting force vectors, acoustic emission energy characteristics, and dynamic tomographic image streams is performed, along with multi-condition joint logic judgments to avoid damage to the dural sac. A six-DOF tool center point motion trajectory is generated based on a trajectory planning algorithm, satisfying the robotic arm's joint constraints, obstacle avoidance requirements, and cutting efficiency, significantly reducing surgical risks.

[0103] Figure 4 A schematic diagram of the frame of the sensing device for autonomous spinal laminar ablation and its immediate stop upon cutting through is shown according to an embodiment of the present invention. The sensing device for autonomous spinal laminar ablation and its immediate stop upon cutting through includes:

[0104] The intraoperative three-dimensional structure perception module 410 is used to perform real-time three-dimensional volume scanning of the target lamina region using an optical coherence tomography probe installed on the end effector of the surgical robot to obtain an initial lamina structure dataset. The initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone mineral density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel.

[0105] The safe cutting region planning module 420 is used to input the initial lamina structure dataset into the lamina bone segmentation network model, and output the lamina bone tissue segmentation results and the probability distribution map of the safe termination surface of the dura mater; based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images and intraoperative scan images, a millimeter-level cuttable feasible region is constructed; the probability distribution map of the safe termination surface of the dura mater is used as the insurmountable safety boundary of the millimeter-level cuttable feasible region, and a six-degree-of-freedom tool center point motion trajectory that meets the requirements of robot arm joint limit, obstacle avoidance and cutting efficiency is generated according to the trajectory planning algorithm to obtain the cutting path planning data;

[0106] The cutting state perception and penetration discrimination module 430 is used to perform autonomous cutting operation under impedance control according to the cutting path planning data; wherein, during the cutting process, real-time perception data stream including real-time cutting force vector, acoustic emission energy characteristics and dynamic tomographic image stream is collected in real time; multi-condition joint logic judgment is performed according to the real-time perception data stream, and the penetration judgment flag signal is triggered according to the judgment result;

[0107] The retraction execution module 440 is used by the underlying motion controller to immediately interrupt the current trajectory interpolation process, stop the feed motion of the cutting tool, and automatically retract along the local normal direction of the vertebral plate in response to the cut-through determination flag signal.

[0108] Figure 5 The diagram shows a structural schematic of an embodiment of the computer device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0109] like Figure 5 As shown, the computer device may include: a processor 502, a communications interface 504, a memory 506, and a communications bus 508.

[0110] The processor 502, communication interface 504, and memory 506 communicate with each other via communication bus 508. Communication interface 504 is used to communicate with other network elements such as clients or other servers. Processor 502 executes program 510, specifically performing the relevant steps in the above-described embodiment of the perception method for autonomous spinal lamina ablation and its stopping upon complete cut.

[0111] Specifically, program 510 may include program code that includes computer operation instructions.

[0112] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0113] Memory 506 is used to store program 510. Memory 506 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0114] According to the solution provided by the present invention, a real-time three-dimensional volumetric scan of the target lamina region is performed using an optical coherence tomography (OCT) probe mounted on the end effector of a surgical robot to obtain an initial lamina structure dataset. This initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel. The initial lamina structure dataset is input into a lamina bone segmentation network model, which outputs the lamina bone tissue segmentation results and a probability distribution map of the safe termination surface of the dural sac. A millimeter-level cuttable feasible region is constructed based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images, and intraoperative scan images. The safe termination surface of the dural sac is then used as the final step. The probability distribution map serves as an insurmountable safety boundary for the millimeter-level cutable feasible area. A six-degree-of-freedom tool center point motion trajectory, satisfying the requirements of robotic arm joint limits, obstacle avoidance, and cutting efficiency, is generated based on the trajectory planning algorithm, resulting in cutting path planning data. Autonomous cutting operations under impedance control are executed based on this cutting path planning data. During the cutting process, real-time sensing data streams, including real-time cutting force vectors, acoustic emission energy characteristics, and dynamic tomographic image streams, are acquired. Multi-condition joint logic judgments are performed based on these real-time sensing data streams, triggering a cut-through judgment flag signal based on the judgment result. The underlying motion controller, responding to the cut-through judgment flag signal, immediately interrupts the current trajectory interpolation process, stops the cutting tool's feed motion, and automatically retracts along the local normal direction of the vertebral lamina. This invention significantly reduces surgical risks by setting a probability distribution map of the dural sac's safe termination surface as an insurmountable safety boundary and generating a six-degree-of-freedom tool center point motion trajectory. Specifically, optical coherence tomography (OCT) and a vertebral lamina bone segmentation network model are used to accurately identify the boundary between the vertebral lamina bone tissue and the dural sac, improving the safety and accuracy of the surgery. During the cutting process, real-time acquisition of data including cutting force vectors, acoustic emission energy characteristics, and dynamic tomographic image streams is performed, along with multi-condition joint logic judgments to avoid damage to the dural sac. A six-DOF tool center point motion trajectory is generated based on a trajectory planning algorithm, satisfying the robotic arm's joint constraints, obstacle avoidance requirements, and cutting efficiency, significantly reducing surgical risks.

[0115] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination of all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed can be employed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose. Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims that list several devices, several of these devices may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be construed as limiting the order of execution.

Claims

1. A method for autonomous cutting of the spinal lamina and a sensing method for stopping the cutting upon completion, characterized in that, include: An optical coherence tomography probe mounted on the end effector of a surgical robot is used to perform real-time three-dimensional volumetric scanning of the target lamina region to obtain an initial lamina structure dataset. The initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone mineral density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel. The initial lamina structure dataset is input into the lamina bone segmentation network model, which outputs the lamina bone tissue segmentation results and the probability distribution map of the safe termination surface of the dura mater. Based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images, and intraoperative scan images, a millimeter-level machinable feasible region is constructed. The probability distribution map of the safe termination surface of the dura mater is used as the insurmountable safety boundary of the millimeter-level machinable feasible region. Based on the trajectory planning algorithm, a six-degree-of-freedom tool center point motion trajectory that meets the requirements of robot arm joint limit, obstacle avoidance, and cutting efficiency is generated to obtain cutting path planning data. The autonomous cutting operation under impedance control is executed according to the cutting path planning data; wherein, during the cutting process, real-time sensing data streams including real-time cutting force vector, acoustic emission energy characteristics and dynamic tomographic image stream are collected in real time; multi-condition joint logic judgment is performed based on the real-time sensing data stream, and the cutting through judgment flag signal is triggered according to the judgment result; In response to the cut-through determination flag signal, the bottom motion controller immediately interrupts the current trajectory interpolation process, stops the feed motion of the cutting tool, and automatically retracts along the local normal direction of the lamina.

2. The method for autonomous cutting of the spinal lamina and its sensing method for stopping cutting upon penetration, as described in claim 1, is characterized in that... Real-time three-dimensional volumetric scanning of the target vertebral lamina region is performed using an optical coherence tomography probe mounted on the end effector of the surgical robot to obtain an initial vertebral lamina structure dataset, which further includes: The lamina-dura mater interface in the surgical field was acquired layer by layer using a frequency domain OCT system, and a high-fidelity data cube was generated by a fast Fourier transform reconstruction algorithm. Each voxel in the data cube carries calibrated bone tissue backscattering reflection intensity information. The local optical signal attenuation coefficient is calculated based on the backscattering reflection intensity information of bone tissue in the data cube. Based on the local optical signal attenuation coefficient, the local signal attenuation characteristics and bone density proxy index that are highly correlated with the microstructure and mineralization degree of bone trabecular bone are derived. The upper and lower cortical boundaries of the lamina were extracted using a surface curvature-guided horizontal set segmentation method, and the lamina thickness distribution was calculated point by point. The strong reflective layer of the epidural membrane was identified by the high-contrast imaging capability of the frequency domain OCT system at the interface between soft and hard tissues. The Euclidean distance field between the strong reflective layer of the epidural membrane and the inner lamina was determined by the subvoxel edge localization method, and an anatomical gap distance atlas between the lamina and the dural sac was generated. The anatomical gap distance atlas includes anatomical gap distance and anatomical structure labels.

3. The method for autonomous cutting of the spinal lamina and its sensing method for stopping cutting upon penetration, as described in claim 1, is characterized in that... The vertebral lamina segmentation network model includes: The voxel embedding layer is used to map the initial lamina structure dataset to a high-dimensional latent space through linear projection, generating an initial voxel embedding tensor with anatomical semantic priors. The pyramid dense encoder layer includes a DenseNet-121 backbone structure, wherein the DenseNet-121 backbone structure adopts a fourth-order downsampling path, each order containing densely connected convolutional blocks and transition layers; it is used to output multi-scale encoded feature maps from coarse to fine based on the initial voxel embedding tensor. The feature decoder layer adopts the nested skip connection structure of U-Net++ to upsample the multi-scale encoded feature map step by step and output the anatomical semantic feature map. The dural sac safe termination surface probability prediction head consists of two parallel branches. The first branch is a fully convolutional integrator classifier, which receives the anatomical semantic feature map and outputs the lamina bone tissue segmentation results. The second branch is an uncertainty-aware regression network, which performs multiple forward propagations under the same input to estimate the prediction variance and outputs a probability distribution map of the dural sac safe termination surface. The probability values ​​of the probability distribution map reflect the risk confidence of the cutting tool reaching the front boundary of the dura mater. The anatomical constraint regularization layer imposes anatomical prior constraints on the outputs of the two parallel branches during the inference phase. Specifically, the minimum gap distance field between the lamina and dura mater obtained by rigid registration of preoperative CT and intraoperative OCT is used to perform morphological clipping and gradient smoothing on the probability distribution map of the safe termination surface of the dural sac, ensuring that the probability peak is strictly located within the inner lamina.

4. The method for autonomous cutting of the spinal lamina and its sensing method for stopping cutting upon penetration, as described in claim 1, is characterized in that... Based on the segmentation results of the vertebral lamina bone tissue, the rigid registration relationship between preoperative computed tomography images and intraoperative scan images, a millimeter-level cuttable feasible region is constructed, which further includes: The preoperative CT image and the intraoperative OCT scan image are spatially aligned using a hybrid rigid registration algorithm, and a six-degree-of-freedom transformation matrix is ​​output. The voxel labels in the segmentation results of the vertebral lamina bone tissue are mapped back to the preoperative CT coordinate system through the inverse transformation of the six-degree-of-freedom transformation matrix to generate the preoperative and intraoperative vertebral lamina bone mask. Using the preoperative and intraoperative lamina bone mask as a base, the safety margin is gradually reduced point by point along the normal vector field of the inner surface of the lamina to generate a conservative cuttable volume; wherein, the safety margin is determined according to the 95% confidence boundary in the probability distribution map of the safe termination surface of the dural sac. Boolean difference operations are performed between the conservative cuttable volume and the key anatomical no-go zones extracted from preoperative CT, including the neural foramen and pedicle, to eliminate potential risk areas and output a millimeter-level cuttable feasible area that simultaneously satisfies the constraints of preserving bone structure integrity and maintaining a safe distance from the dura mater.

5. The method for autonomous cutting of the spinal lamina and its sensing method for stopping cutting upon penetration, as described in claim 1, is characterized in that... Based on the real-time sensed data stream, a multi-condition joint logic determination is performed, and the triggering of the cut-through determination flag signal based on the determination result further includes: Based on real-time sensing data stream, a multi-condition joint logic judgment is performed. When the judgment result simultaneously meets the preset conditions, the cut-through judgment flag signal is triggered. One of the preset conditions is: the real-time cutting force vector amplitude decreases by more than a preset threshold and its direction deviates from the bone surface normal. The second preset condition is: the acoustic emission energy characteristics show an attenuation inflection point and indicate penetration of the bone-dura mater interface; The third preset condition is: the strong reflection layer of the corresponding dural capsule membrane is detected for the first time in the dynamic tomographic image stream and appears continuously for more than three frames at the cutting front, and the distance between it and the current tool position is less than the minimum safe distance of the corresponding position in the dissection gap distance map.

6. The method for autonomous cutting of the spinal lamina and its sensing method for stopping cutting upon penetration, as described in claim 1, is characterized in that... The local signal attenuation characteristics are quantized using an anisotropic attenuation tensor field, which is fused based on the Carter-Hayes bone force model; wherein, the expression for the anisotropic attenuation tensor field is: ； in, Bone mineral density proxy index derived from OCT; , Each with points The cortical bone neighborhood and the cancellous bone neighborhood are centered. , ; , For the relevant length, , ; , These are the squares of the volumes of the cortical bone neighborhood and the cancellous bone neighborhood, respectively. OCT intensity volume data At point Spatial gradient along the i-th coordinate axis; OCT intensity volume data At point The spatial gradient along the j-th coordinate axis; , For spatial variables; Threshold for bone tissue type discrimination; voxel point The first anisotropic decaying tensor field One component; The coordinates are the spatial positions of the inner plate of the lamina.

7. The method for autonomous cutting of the spinal lamina and its sensing method for stopping cutting upon penetration, as described in claim 4, is characterized in that... The expression for the safety margin is: ； in, ; ; This refers to the thickness distribution of the vertebral lamina. The spatial coordinates of the inner plate of the lamina; This is the dividing depth; As the weight core, ; It is a two-dimensional offset vector within the cross section; To correspond to the half-width of the cutting edge of a high-speed drill, ; The confidence coverage threshold. ; Let be a transverse cross-sectional element perpendicular to the normal at depth s; The probability distribution of the safe termination surface of the dural sac is the value of the corresponding spatial location; These are candidate security depth variables.

8. The method for autonomous cutting of the spinal lamina and its sensing method for stopping cutting upon penetration, as described in claim 5, is characterized in that... The penetration confidence level of the multi-condition joint logic decision is: ； in, For reference cutting force, ; Bone mineral density proxy index derived from OCT; The acoustic emission coefficient; This represents the minimum safe gap in the local area derived from the anatomical gap distance map; It is a Gaussian trust function; This is the real-time cutting force vector; The second time derivative of the acoustic emission energy; For indicator functions; The Euclidean distance from the current tool center point TCP to the strong reflective layer of the dural capsule; The number of consecutive detection frames; For time.

9. A sensing device for autonomous cutting of the spinal lamina and stopping upon cutting through, characterized in that, include: The intraoperative three-dimensional structure perception module is used to perform real-time three-dimensional volume scanning of the target lamina region using an optical coherence tomography probe installed on the end effector of the surgical robot to obtain an initial lamina structure dataset. The initial lamina structure dataset includes bone tissue reflection intensity information, local signal attenuation characteristics, lamina thickness distribution, bone mineral density surrogate index, anatomical gap distance between the lamina and the dural sac, and anatomical structure labels for each voxel. The safe cutting region planning module is used to input the initial lamina structure dataset into the lamina bone segmentation network model, and output the lamina bone tissue segmentation results and the probability distribution map of the safe termination surface of the dura mater; based on the rigid registration relationship between the lamina bone tissue segmentation results, preoperative computed tomography images and intraoperative scan images, a millimeter-level cuttable feasible region is constructed; the probability distribution map of the safe termination surface of the dura mater is used as the insurmountable safety boundary of the millimeter-level cuttable feasible region, and a six-degree-of-freedom tool center point motion trajectory that meets the requirements of robot arm joint limit, obstacle avoidance and cutting efficiency is generated according to the trajectory planning algorithm to obtain the cutting path planning data; The cutting state perception and penetration discrimination module is used to perform autonomous cutting operation under impedance control according to the cutting path planning data; wherein, during the cutting process, real-time perception data stream including real-time cutting force vector, acoustic emission energy characteristics and dynamic tomographic image stream is collected in real time; multi-condition joint logic judgment is performed based on the real-time perception data stream, and a penetration judgment flag signal is triggered according to the judgment result; The retraction execution module is used by the underlying motion controller to immediately interrupt the current trajectory interpolation process, stop the feed motion of the cutting tool, and automatically retract along the local normal direction of the vertebral plate in response to the cut-through determination flag signal.

10. A computer device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the autonomous cutting of the spinal lamina and the sensing method of stopping after cutting through, as described in any one of claims 1-8.

Images