Dynamic bone recognition for robotic vertebral plate cutting via unit energy consumption and SVM optimized by PSO

Current orthopedic robots lack the ability to dynamically sense or accurately recognize bone layers during vertebral plate decompression surgery, limiting their ability to adjust actions in real time as skilled surgeons do. This study aims to improve robotic vertebral plate cutting by developing a bone recognition model that utilizes a unit energy consumption feature vector and support vector machines (SVM) optimized with particle swarm optimization (PSO). An experimental setup using fresh pig bones of varying densities was established, and cutting experiments were performed under different parameters. Force signals from various cutting directions were analyzed, and wavelet threshold noise reduction was applied to transverse cutting forces. A feature space distribution was mapped, and total energy consumption was calculated to create the unit energy consumption function. Feature vectors were spatially mapped, and the effectiveness of energy consumption-based feature extraction was assessed. Principal component analysis (PCA) was used for further feature extraction and dimensionality reduction. The data was normalized, and an SVM-based bone identification model was developed, optimized by PSO. The optimized model achieved bone identification accuracy of 90.64%, compared to 83.56% using traditional feature extraction techniques. Cross-validation through experiments demonstrated a 7.08% improvement in classification accuracy. The study confirms the feasibility of the predictive bone recognition model, which enhances the precision of robotic vertebral plate cutting by enabling real-time dynamic adjustment of cutting parameters based on bone type.