Deep-learning-based instrument detection for intra-operative robotic assistance.

Area of Science:

• Robotics
• Computer Vision
• Surgical Technology

Background:

• Robotic scrub nurses offer advanced operating room solutions.
• Accurate surgical instrument detection is crucial for robotic assistance.
• Wisdom teeth extraction requires a specific set of surgical instruments.

Purpose of the Study:

• To develop and evaluate a data augmentation technique for surgical instrument detection.
• To improve the training efficiency of deep learning models for robotic scrub nurses.
• To create a comprehensive dataset for wisdom teeth extraction instrument detection.

Main Methods:

• A dataset of 369 multi-instrument images was created using a robotic scrub nurse system.
• The Mask-Based Object Insertion method was proposed for synthetic data generation.
• Mask R-CNN models were trained and evaluated using real and augmented data.

Main Results:

• Models trained with synthetic data achieved performance comparable to models trained with real images.
• Combining real and synthetic data improved model generalization.
• The data augmentation technique significantly reduced manual labeling efforts.

Conclusions:

• The proposed data augmentation method effectively reduces labeling workload for AI-driven surgical tools.
• A publicly available dataset for wisdom teeth extraction instruments is provided.
• The research facilitates the development of more sophisticated robotic surgical assistance systems.