Real-time surgical instrument tracking in robot-assisted surgery using multi-domain convolutional neural network

Area of Science:

• Robotics
• Computer Vision
• Medical Imaging

Background:

• Real-time surgical instrument tracking is crucial for computer-assisted interventions in robot-assisted surgery.
• Accurate tracking enables visual and haptic feedback for enhanced surgeon-robot interaction and precise control.

Purpose of the Study:

• To develop a fast and accurate 2D surgical instrument tracking method for robot-assisted surgery.
• To address the limitations of existing tracking datasets by introducing a new, comprehensive dataset.

Main Methods:

• Application of a multi-domain convolutional neural network for 2D surgical instrument tracking.
• Utilization of a focal loss function to mitigate the impact of easily classifiable negative examples.
• Development of a new dataset incorporating m2cai16-tool data and cadaver experiments.

Main Results:

• The proposed multi-domain convolutional neural network achieves fast and accurate 2D surgical instrument tracking.
• The method demonstrates superior performance compared to state-of-the-art real-time trackers on the newly introduced dataset.
• The focal loss effectively reduces the influence of simple negative samples, improving tracking robustness.

Conclusions:

• The developed method offers a significant advancement in real-time surgical instrument tracking for robot-assisted surgery.
• The new dataset provides a valuable resource for evaluating and advancing surgical tool tracking algorithms.
• This work contributes to the development of more sophisticated and reliable computer-assisted surgical systems.