Core Concepts
A computer vision algorithm can accurately identify a wide range of surgical instruments, with potential to optimize surgical tray management, prevent instrument loss, and quantify instrument usage.
Abstract
This study aimed to develop a computer vision (CV) algorithm to accurately identify and classify surgical instruments commonly used in neurosurgery. The researchers collected a dataset of 1,660 images of 27 different neurosurgical instruments, labeled them using bounding boxes, and trained a U-Net convolutional neural network model to perform pixel-level classification of the instruments.
The key findings are:
- The U-Net model achieved 80-100% accuracy in identifying 25 different instrument classes, with 19 out of 25 classes having over 90% accuracy.
- The model had lower accuracy (60-80%) in sub-classifying certain similar-looking forceps (Adson, Gerald, Debakey).
- The intersection-over-union (IoU) scores, which measure pixel-level accuracy, ranged from 0.4263 to 0.8566 across the different instrument classes.
- The researchers demonstrated the potential of using computer vision to track surgical instruments, optimize surgical tray management, prevent instrument loss, and quantify instrument usage during procedures.
The authors note that more training data, especially from real operating room conditions, will be needed to increase the accuracy across all surgical instruments. Integrating this technology into the operating room could lead to improved efficiency, cost savings, and patient safety by automating instrument tracking and management.
Stats
Accuracy of instrument identification ranged from 63.64% for Adson Forceps to 100% for Irrigation Bulbs.
19 out of 25 instrument classes had over 90% accuracy.
Intersection-over-union (IoU) scores ranged from 0.4263 for Tonsil Forceps to 0.8566 for Irrigation Bulb.
Quotes
"We demonstrated the viability of using machine learning to accurately identify surgical instruments. Instrument identification could help optimize surgical tray packing, decrease tool usage and waste, decrease incidence of instrument misplacement events, and assist in timing of routine instrument maintenance."
"Such technology and its data stream have the potential to be used as a method to track surgical instruments, optimize data around instrument usage and instrument supply in the operating room, evaluate surgeon performance, help with instrument inventory and organization, prevent incidents such as retained foreign objects, and quantitatively describe how to do more with less."