Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Device Technologies and Biomedical Robotics

Device Technologies and Biomedical Robotics - Poster Session C

Poster I17 - Machine Learning Meets Sterile Environments: Validation of Motion Capture Systems for Evaluating Medical Device Package Handling

Friday, October 25, 2024
10:00 AM – 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  In healthcare settings, surgical site infections (SSIs) are a major risk to patient safety and a burden to the healthcare system [5, 9]. SSIs are defined as an infection related to a surgical procedure that occurs near the surgical site within 30 days or 90 days after surgery, depending on the type of procedure [3]. SSIs can be caused by indirect contamination of medical devices during routine opening and handling of medical device packaging [10].

Currently, medical device manufacturers validate package usability and handling using experimental strategies to meet U.S. and European medical device regulations using international standards [4,7,8]. However, manufacturers lack critical baseline data [1], and there is a need for objective measures of human interactions and handling of packaging [1, 10]. A quantitative method for analyzing medical device package handling could inform altered medical device package handling procedures and improved package designs, targeting a reduction in the rate at which SSIs occur.

The purpose of this study was to develop and validate a method to measure hand motions during medical device package handling. To accomplish this goal, two motion-capture systems were used to evaluate known translations of a hand model and define the measurement volume. It was hypothesized that the accuracy of the two systems would be equivalent under the laboratory test conditions.

Materials and

Methods:  The two motion-capture systems used were a reflective marker system and a machine-learning framework. The reflective marker system (Polaris Vicra) is a two-camera system that is based on external marker arrays on objects of interest and has a 95% confidence interval of 0.5mm in translational motions [6]. The machine-learning framework (Mediapipe) uses a standard video camera feed and incorporates a machine-learning framework that does not require external markers [2].

A custom test rig was fabricated to enable controlled translations of a mounted 3D-printed hand model (thingiverse.com/thing:1286) along three orthogonal axes (Figure 1). The model was placed at an initial working distance of 1.524 meters from the end of the lens of the machine-learning camera. The rig enabled translations of 142 mm, 115 mm, and 609.6 mm along the X, Y, and Z axes, respectively. Calibration of the machine-learning system was completed by moving the model a known distance in the calibration plane in both the x and y axes separately, creating a baseline coefficient for translating between media pipe units and millimeters in both axes. The two systems were tested at the boundaries of the measurement volume, which was defined using the edges of the visible range of the camera system. The model was then tested at the positive and negative extremes of each axis. The accuracy was determined for each system using the accuracy equation (Figure 2), which compares the absolute difference between the measured translation and the known translations in each axis, with 100% representing perfect accuracy.

Results, Conclusions, and Discussions: The accuracy of the machine-learning system was generally less than that of the marker-based motion-capture system, 85% compared to >99%, respectively. The testing volume had dimensions of 115 mm height, 142 mm width, and 610 mm of depth corresponding to approximately 10x106 cubic mm of space. Generally, the accuracy of the marker-based motion-capture system was consistent throughout the testing volume, whereas the accuracy of the machine-learning system declined for out-of-plane translations moving toward the camera (+Z-axis). For example, the accuracy of the machine-learning system was at least 98% accurate for X-axis and Y-axis translations near the calibration plane and behind it (-Z axis), but dropped to 85.7% accuracy for X-axis and Y-axis translations in front of it (+Z axis). But, only considering the performance of the machine-learning system in or behind its plane of calibration during the experiments, both motion-capture systems performed within 2% accuracy of each other.

Applying motion capture techniques to an optical measurement tool using Mediapipe can greatly increase potential applications in an operating room setting without the need for placing reflective markers that are needed in traditional marker-based motion capture systems. While the machine-learning-based system is generally more applicable in those settings, situations that require a higher degree of accuracy would benefit from the use of marker-based motion-capture systems. Additionally, the information from the machine-learning-based system takes more time to process and future work will involve optimizing post-processing of the video feed. Additionally, future work for this project aims to develop a quantitative scale for measuring the efficiency of opening medical device packaging by tracking hand movements and packaging interactions. Ultimately, the goal is to define package handling that can reduce the potential for contamination of sterile medical devices and the incidence of SSIs.

Acknowledgements (Optional):  We acknowledge the Clemson University Creative Inquiry Program for funding and administrative support.

References:

1. Bix, L., Zhang, Y., & Trier, T. (2020). Unpackaging Aseptic Presentation: A Qualitative Study into the Contextual Influences Involving Medical Packaging Design and Use Heuristics Among Perioperative Personnel. Journal of Applied Packaging Research, 12(1).

2. Google. (2024, May 21). Hand landmarks detection guide  |  Google AI Edge  |  Google for Developers. Google. https://ai.google.dev/edge/mediapipe/solutions/vision/hand_landmarker

3. Horan, T. C., Andrus, M., & Dudeck, M. A. (2008). CDC/NHSN surveillance definition of health care-associated infection and criteria for specific types of infections in the acute care setting. American journal of infection control, 36(5), 309–332. https://doi.org/10.1016/j.ajic.2008.03.002

4. ISO 11607-1 and 2:2019 - Packaging for terminally sterilized medical devices -- Part 1: Requirements for materials, sterile barrier systems and packaging systems. Part 2: Packaging requirements for forming, sealing, and assembly processes.

5. Owens, C. D., & Stoessel, K. (2008). Surgical site infections: Epidemiology, Microbiology and Prevention. Journal of Hospital Infection, 70 (suppl 2), 3–10. https://doi.org/10.1016/s0195-6701(08)60017-1

6. Polaris Vicra: Compact Performance for MED device oems. NDI. (2023, November 29). https://www.ndigital.com/optical-navigation-technology/polaris-vicra/#:~:text=The%20Polaris%20Vicra%20tracks%203D,confidence%20interval%20of%200.5%20mm.

7. Quality System Regulations, 21 C.F.R. § 820 (2024)
https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr=820.75

8. Regulation (EU) 2017/745 of the European Parliament and of the Council of 5 April 2017 on medical devices, amending Directive 2001/83/EC, Regulation (EC) No 178/2002 and Regulation (EC) No 1223/2009 and repealing Council Directives 90/385/EEC and 93/42/EEC Official Journal of the European Union, Volume 60: L 117, 5 May 2017. P 1-175. Retrieved from http://data.europa.eu/eli/reg/2017/745/2023-03-20 on July 25, 2024)

9. Urban, J. A. (2006). Cost analysis of surgical site infections. Surgical Infections, 7(s1). https://doi.org/10.1089/sur.2006.7.s1-19

10. Vocelle, A. R., Trier, T., Bix, L., & Bush, T. R. (2019). A method for quantifying key components of the opening process for opening pouch-style packages containing medical devices. Applied ergonomics, 76, 97–104. https://doi.org/10.1016/j.apergo.2018.12.010