markmamed/imu-surgical-intention-perception

This Python project is a machine learning system designed to recognize a surgeon's intentions during an operation by reading motion data from IMU sensors, small devices that measure movement, acceleration, and rotation. The goal is to understand what surgical gesture or task a surgeon is performing in real time, which could feed into a robotic surgical assistant or exoskeleton that adapts to the operator's next move. The system is trained on several publicly available motion datasets including JIGSAWS, which contains recordings of surgical tasks like knot tying, needle passing, and suturing. It extracts features from the sensor data, such as speed, acceleration, and how both hands move together, and trains machine learning classifiers to predict which surgical subtask is currently happening. The README reports that the best model combination currently reaches about 84% accuracy on held-out test data. The project includes tools for training, running predictions on new CSV sensor files, and generating charts that visualize how the model performs and how a surgeon's activity transitions over time. It is a research codebase from Peking University Health Science Center.