Enhanced 2D Hand Pose Estimation for Gloved Medical Applications: A Preliminary Model.

Area of Science:

• Medical technology
• Computer vision
• Machine learning

Background:

• Accurate medical-gloved hand tracking is increasingly vital for assessing and training healthcare practitioners.
• Evolving digital health technology necessitates tools to reduce procedural errors in clinical environments.

Purpose of the Study:

• To develop and validate a computer vision model for accurate hand pose estimation of medical-gloved hands.
• To assess the model's efficacy in tracking skeletal hand movements during aseptic drug compounding.

Main Methods:

• Utilized computer vision and machine learning (DeepLabCut) for hand pose estimation.
• Recorded high-definition video of practitioners performing aseptic drug compounding while wearing medical gloves.
• Trained and tested the model using an 80/20 split with manual annotation of hand poses.

Main Results:

• The model achieved an average root mean square error (RMSE) of 5.89 pixels on training data and 10.06 pixels on test data.
• Excluding low-confidence keypoints improved the test set RMSE to 7.48 pixels, indicating high tracking accuracy.
• Demonstrated effective hand movement tracking in both controlled and in situ drug compounding scenarios.

Conclusions:

• The developed hand pose estimation model offers a pioneering method for tracking medical-gloved hands.
• This technology has significant potential for enhancing clinical training and ensuring procedural safety, especially in high-precision tasks like drug compounding.