Imaging Navigation And State Recognition In Robot-assisted Decompressive Laminectomy

Decompressive laminectomy is a very important way to treat some orthopedic diseases such as spinal stenosis.Surgeons should hold instruments for a long time to cut the lamina carefully,which causes physical fatigue,and there is a chance that the dural tears and some other adverse conditions occur,leading to cerebrospinal fluid leakage or hematoma which forms pressure on spinal nerves.Surgical robot has the characteristics of high precision,high reliability,easy interaction,no fatigue,etc.which can assist surgeons to complete fine decompressive laminectomy under their plans,improving operation efficiency and reducing incidence of complications.In spine surgery,the robot mainly completes the insertion of pedicle screw based on 2D image planning and navigation assistance,which mainly realizes the positioning function guided by navigation.At present,there is no planning method for decompressive laminectomy.The lamina of spine is a free-form surface in 3D image space,which exist incompletely reconstructed surface,some interference with interaction and other issues.When the robot clamps the piezosurgery for cutting the bone,the physiological movement of the spine caused by breathing will produce additional positive pressure,which affects the cutting stability.And when the robot clamps the high-speed grinding drill for grinding the bone,the convected motion of the spine caused by external force will produce additional displacement s,which may affect the grinding accuracy.The robot has no similar adaptability to surgeons for this dynamic environment.The internal conditions of vertebrae can not be observed in real time through endoscopy during decompressive laminectomy.It is necessary to recognize the operation state through online analysis of sensor signals,and then provides early warning or emergency stop for operation,so as to improve the safety of robot-assisted surgery.However,the accuracy of current state recognition is not good and lacks a unified criterion.The medical imaging is enhanced based on deep learning,and the quality of reconstructed image and the interaction way of lesion area are optimized,and both the path and speed planning methods for decompressive laminectomy are designed.Multi-label semantics segmentation of the CT image of the vertebrae based on the U-Net derivative network is used to realize extraction of target lamina and local reconstruction.The interactive planning between surgeons and reconstructed image is realized by the space guide line and bounding box,which can help accomplish the initial point location,variable speed grinding and constraints construction,etc.The relationship model between ventilator output and fluctuation in the spine is established,and an additional force compensation method is proposed based on elastic fixture and fuzzy control.Combining with the actual parameters of clinical ventilator and its pipeline’s parameters,the gas flow state under different breathing phases is analyzed,and the mathematical model between the gas pressure and flow at the outlet of the ventilator and the amount of spine fluctuation is established.By designing the elastic clamping tool and the fuzzy stiffness controller,the robustness of piezosurgery to the cutting environment is improved.The bending and torsion model of the spine is established based on traditional elastic mechanics,and the method for parameter identification and the displacement compensation strategy for clinical application are proposed.The relationship model between the force and its angle actting on the lamina and the displacement of the local vertebral area in the direction of sagittal axis and coronal axis is established.The robot controller is designed based on the adaptive Kalman filter with unknown statistical characteristics of noise.The simulation is carried out based on PD control and two-link robot to verify the dynamic performance.The real-time standardized features of intraoperative sound and force signals are constructed for state recognition,including the vertebral cortical bone,transition area,cancellous bone,etc.The time-frequency analysis of sound signal is carried out based on FFT.The features of sound signal are defined by constructing the regression equation in frequency domain,which is used to recognize the transition area.The features of force signal are improved by considering surgeon’s perception on the state,which can be used to recognize the cortical bone and transition area online.The two kinds of signals are fused based on D-S evidence theory,and the state recognition of surgery is improved.At present,the research of surgical robot at home and abroad mainly focuses on insertion of pedicle screw,which is not yet mature for decompressive laminectomy.The technologies of robot-assisted decompressive laminectomy are systematically studied.Combining deep learning,computer vision,mathematical modeling and simulation,information fusion and other technologies,the issues,such as imaging planning,automatic registration,dynamic compensation and state recognition,are focused on solving,which can provide technical support for the application of the robot-assisted decompressive laminectomy.