Surgical Gesture Recognition in Robot-Assisted Surgery Using Machine Learning Methods on Kinematic Data.

Area of Science:

• Robotics
• Machine Learning
• Surgical Technology

Background:

• Robot-assisted surgery offers precision but lacks real-time intraoperative feedback.
• Accurate recognition of surgical gestures is crucial for developing intelligent surgical tools.
• Current methods often rely on visual data, limiting real-time kinematic analysis.

Purpose of the Study:

• To develop and evaluate machine learning models for real-time surgical gesture recognition.
• To utilize exclusively kinematic data from patient-side robotic manipulators.
• To improve upon existing state-of-the-art recognition rates for surgical tasks.

Main Methods:

• Training various neural network architectures, including a Long Short-Term Memory (LSTM) baseline.
• Proposing and testing two hybrid models: LSTM with Conditional Random Field (CRF) and LSTM with an attention layer.
• Evaluating model performance on the JIGSAWS dataset, focusing on suturing tasks.

Main Results:

• The proposed hybrid approaches outperformed the baseline LSTM model.
• The LSTM model combined with an attention layer achieved the highest accuracy at 81.56%.
• Comparative analysis identified specific areas for further performance optimization.

Conclusions:

• Hybrid machine learning models, especially attention-based ones, show significant promise for real-time surgical gesture recognition.
• Kinematic data alone is sufficient for effective gesture recognition in robot-assisted surgery.
• This research provides a foundation for developing intraoperative monitoring and assistance tools for surgeons.