Development of an AI solution for surgical gauze management

• Main Author: Ding, Jishen
• Other Authors: Cai Yiyu
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Mechanical engineering::Surgical assistive technology
• Online Access: https://hdl.handle.net/10356/167017
• Institution: Nanyang Technological University

Surgical procedures often require the use of gauze to prevent bleeding and infections, but manual gauze counting can be time-consuming, prone to error, and pose a potential threat to patient safety if any gauze is left inside a patient. This project developed an automated system for detecting and counting surgical gauze during surgery to address the challenge. Two hardware frames were designed to hold the camera and processor to ensure the system's stability and usability in the operation theatre environment. The development of automatic counting software and user interface, which eliminated the need for human input. In addition, a human detection function was integrated into the system to prevent human error. Over 1000 raw data points were collected, and three representative AI models (GauzeV5, GAUZEBW, GAUZECOLOR) were trained based on the YOLOv5 object detection algorithm discussed in this report. The models were evaluated using precision, recall, F1 score, and mAP@0.5, with GauzeV5 performing well in detecting clean gauze, but failing to detect blooded gauze in real-world testing. GAUZEBW and GAUZECOLOR models performed better in detecting bloodied gauze but at the cost of sacrificing some confidence levels in detecting clean gauze. Overall, the choice of model depends on the specific application condition and the trade-off between detection accuracy and computational resources. This project demonstrates the feasibility of a gauze-counting product that can improve surgical efficiency and reduce the risk of leaving gauze inside patients. The proposed system provides a foundation for the development of a reliable gauze management product for use in real-world surgical settings. By automating the gauze counting process, the system can potentially save time and reduce the risk of human error. Future works are suggested at the end of this report.