Attention-Based CNN-LSTM for Enhanced Perception of Bone Milling States in Surgical Robots

Abstract

Bone milling surgery is one of the commonly used methods for the treatment of orthopedic diseases in clinics, and the use of robots to assist in the completion of bone milling surgery can effectively improve the efficiency of the operation. However, since the addition of robots will diminish the surgeon’s ability to perceive the surgical environment, improving the robotic system’s ability to perceive the milling state is an important factor in improving surgical safety. Inspired by the human auditory perception mechanism, this article proposes a method to recognize the milling state (including the type, depth, and angle of milling bones) using milling acoustic signals combined with neural networks. In this study, a series of bone milling experiments were completed to construct the datasets. The correlation between acoustic signal features and the milling state is analyzed, and then 17-D features are extracted as inputs to the network model. A novel auditory attention model combining the attention mechanism with the CNN-LSTM network for bone milling state recognition has been proposed, and the effects of different network structures on recognition accuracy are discussed. The results indicate that the proposed classification network has a great recognition accuracy for milling states, which proves that the method is beneficial for improving the safety of automatic milling surgery for robots.