Real-Time Surgical Tool Classification and Characterization with Surface-EMG Signals

The selection and utilization of micro-surgical in-struments play a crucial role in determining the outcome of modern surgical procedures. The surgeon's skill in using these tools significantly influences the success or failure of the surgery. Common micro-surgical instruments like forceps, clamp applicators, micro-scissors, and needle holders are integral to these procedures, and understanding their use in conjunction with the surgeon's expertise is an ongoing area of interest. While various computer vision-based methods have been developed to identify and segment these instruments in surgical settings using microscopic recordings, they often lack the ability to capture the surgeon's haptic feedback and subtle variations in tool handling. Video-based detection systems are also prone to issues like masking and shadowing, which can obscure crucial details during surgery. Therefore, there is a need to extract localized information during surgery in addition to information obtained from video recordings. Surface electromyography (sEMG) signals, which capture electro-muscular activity on the skin, offer valuable insights into the actions performed during surgery. This study explores the potential of using sEMG signals to detect and classify micro-surgical instruments. A specialized two-channel sEMG acquisition system, connected to electrodes, was designed to create a carefully staged and supervised dataset involving the use of surgical instruments. Five distinct features were derived from each channel and utilized to develop a highly accurate model. Through the use of a Convolutional Neural Network (CNN) model, an impressive accuracy of 92.8% was achieved during the training phase, and real-time classification further demonstrated an accuracy of 92%.