Efficient computer vision pipeline for automated anesthetic injection documentation.

Area of Science:

• Medical technology
• Computer vision
• Anesthesiology

Background:

• Accurate documentation of anesthetic injections is crucial for patient safety and procedural efficiency.
• Current manual documentation methods are prone to errors and can be time-consuming.
• Variability in stopcock placement and syringe sizes presents challenges for automated systems.

Purpose of the Study:

• To develop and validate a novel computer vision pipeline for automated documentation of anesthetic injection events.
• To enhance the accuracy and reliability of recording injection events and administered anesthetic volumes.
• To address challenges such as stopcock placement variability and different syringe sizes.

Main Methods:

• A computer vision pipeline utilizing the Segment Anything Model (SAM) for syringe segmentation.
• Vector similarity matching for generalization across syringe sizes and occlusions.
• Integration of lightweight methods for motion detection, syringe classification, and quasi-real-time volume estimation.

Main Results:

• 100% sensitivity in detecting injection events.
• An overall documentation success rate of 86.3%.
• Mean absolute errors (MAE) for volume estimation ranged from 0.10 ml (3ml syringes) to 0.61 ml (20ml syringes), outperforming manual measurements.

Conclusions:

• The developed computer vision pipeline offers a fully automated, camera-only solution for anesthetic injection documentation.
• The system demonstrates quasi-real-time performance, suitable for clinical workflow integration.
• This approach significantly improves documentation accuracy, standardizes procedures, enhances patient safety, and reduces anesthesiologist workload.