All-Encompassing Bone Recognition and Safety Control Strategy for Robot-Assist Laminectomy Based on Milling Force

The incidence of spinal degenerative diseases is increasing yearly, due to the global population aging. Spinal surgery, such as laminectomy, remains the most effective, comprehensive, and widely practiced treatment for spinal degenerative disease. As the lamina is adjacent to the spinal cord, spinal nerve injury is one of the most common complications of freehand laminectomy. In view of the high risk, complexity, and extensive learning curve of laminectomy, the surgical robots can improve the safety of surgical operation and have been gradually applied in orthopedics. However, robotic automated laminectomy was challenging because accurate state recognition and safety control are the prerequisites for robotic control, which lack of mature research. Accurate recognition of the inner cortical bone and precise robotic control are key to ensure that lamina is not pierced, which can prevent spinal nerve injury. Therefore, we took the lead in proposing a bone recognition model based on extreme learning machine (ELM) to recognize bone substance for various lamina milling conditions effectively. Specifically, the proposed model captures the features of milling force in spatial order data to produce richer bone characteristics representations. Also, based on our model, a control strategy was proposed to ensure robots stop timely when the milling instruments contact with the inner cortical bone. We verified the effectiveness of our recognition model and control strategy by in vivo and in vitro experiments and obtained great bone recognition performance. Notably, the recognition accuracy reached 100% under most milling conditions. And the laminectomy conducted by robot-assist automatic was controlled more safely than freehand (P value $2.2110-5$ and $1.7510-5$ ).