An Experimental Study Of Multimodal Information Perception In Spinal Surgery Area And Its Application In Robot-assisted Surgery

Background:The annual incidence of spinal diseases,which are represented by cervical spondylosis,spinal stenosis,and intervertebral disc herniation,has recently been on the rise.These diseases are the leading causes of neck and shoulder discomfort,low back pain,limb pain,and intermittent claudication.Spinal surgery is the most effective treatment to relieve spinal cord or nerve root compression and alleviate the related clinical symptoms.However,spinal surgery involves several complications,such as high operative difficulty and risk,various complications,and long training time.Effective reduction of the difficulty,risk,and postoperative complications associated with spinal surgery needs drastic attention in orthopedics.Compared with traditional manual operation,robot-assisted surgery offers higher freedom of movement and procedural accuracy,thereby improving safety and effectiveness.Therefore,it has been applied and promoted in many clinical departments,including orthopedics.Currently,spinal surgery robots primarily help navigate in the surgical field to assist pedicle screw placement.There are no mature operational robots aimed at bone grinding due to the lack of multimodal information perception and feedback of the surgical field.In manual surgery,surgeons mainly use their tactile sensations to perceive and judge the state of bone grinding combined with their own experience during the operation.However,this method has obvious problems,such as strong subjectivity,poor stability,and inconsistent standards.Therefore,perception of tissue information in the operative area during spinal surgery,simulation of judgment and decision-making process of surgeons,and achieving a safe and accurate grinding operation are the core hurdles affecting the development of spinal surgery robots.Objective:The present study aimed to develop a multimodal perception of different tissues in the surgical field during spinal surgery,establish a state recognition method and safety control strategy for vertebral lamina grinding,and verify the accuracy and effectiveness of robot-assisted lamina grinding.Methods:Acoustics and force sensing was first used to collect the signals during the grinding process of the spine based on a biomedical engineering experimental platform.Subsequently,filter noise reduction and feature analysis were performed to determine the characteristic values and change rules of signals under different grinding parameters.Next,bioelectrical impedance sensing was applied to measure the electrical impedance values of different tissues in the spinal surgery field,and the feasibility of tissue identification was verified using statistical analysis.Based on multimodal perception,the backpropagation(BP)neural network was used by combining data on various grinding parameters,surgical instruments,and grinding forces of the bone layer to establish a state recognition model and safety control strategy for the robot-assisted lamina grinding.Finally,the accuracy of the state recognition model and the effect of robot-assisted lamina grinding were evaluated by conducting a live surgery on experimental pigs.Results:The results of multimodal perception showed that the force sensing technique could effectively identify the different bone tissues and different grinding parameters by extracting the characteristic values of the force signal and analyzing the control variable.The grinding force perception was suitable for a robot-assisted spinal operation.By analyzing the short-term energy of acoustic signals,acoustic sensing helped recognize the different types of bone tissues.Subsequently,the bioelectrical impedance spectrum of different tissues in the spinal surgery area was constructed,and the distinction between different types of bone tissues and non-bone tissues was achieved.Experiments on the state recognition model for lamina grinding showed that the BP neural network could help recognize the bone layer and the judgment of the grinding end,which used the grinding parameters and force characteristic values as input and gave the bone layer information as the output data.The results from animal experiments confirmed that the accuracy of the state recognition method and the safety control strategy could reach beyond 90%,and the grinding degree of the vertebral lamina was approximately 90%.Conclusion:Multimodal perception based on sensing force,acoustic,and bioelectrical impedance signals could help identify different tissues in the spinal surgery field.Meanwhile,the state recognition method and safety control strategy based on the BP neural network can facilitate accurate,stable,and efficient lamina grinding during robot-assisted surgery.