Spine endoscopic atlas: an open-source dataset for surgical instrument segmentation

Area of Science:

• Neurosurgery
• Medical Imaging
• Artificial Intelligence

Background:

• Endoscopic spine surgery (ESS) offers minimally invasive benefits but faces adoption challenges due to a steep learning curve.
• Artificial intelligence (AI) is vital for improving ESS precision and safety.
• Accurate segmentation of surgical instruments is essential for AI-driven surgical assistance.

Purpose of the Study:

• To introduce the Spine Endoscopic Atlas (SEA) dataset for training AI models in ESS.
• To facilitate precise instrument segmentation in endoscopic spine surgery.

Main Methods:

• Creation of the Spine Endoscopic Atlas (SEA) dataset with 48,510 annotated images.
• Inclusion of 10,662 instrument segmentations from real-world ESS procedures.
• Validation of five deep learning models using the SEA dataset.

Main Results:

• The SEA dataset enables improved segmentation accuracy for surgical instruments in complex ESS scenarios.
• Demonstrated value of the dataset in enhancing deep learning model performance.
• Established a foundation for AI advancements in endoscopic spine surgery.

Conclusions:

• The Spine Endoscopic Atlas (SEA) dataset is a valuable resource for developing AI tools in ESS.
• This dataset supports the advancement of intelligent surgical assistance systems.
• SEA facilitates enhanced precision and safety in minimally invasive spinal surgery.