Deep Learning-Based Integrated System for Intraoperative Blood Loss Quantification in Surgical Sponges

Accurate quantification of intraoperative blood loss is crucial for enhancing patient safety and the success rate of surgeries. Traditional estimation techniques, mainly reliant on visual assessments, are prone to significant inaccuracies due to their subjective nature. This study introduces MDCare, an innovative deep learning-integrated system designed to substantially improve the precision of blood loss quantification using surgical sponges. By integrating advanced hardware components, including a mass sensor and webcam, with sophisticated algorithms like ResNet-18 and YOLOv4, MDCare achieves classification accuracy up to 96.2% and sponge detection accuracies above 91% for both synthetic and real blood scenarios. The system processes images at 7.4 frames per second, aligning with the exigent pace of surgical environments, thereby supporting surgeons with real-time, accurate blood loss data essential for timely and informed decision-making. The contributions of the paper are: (1) Demonstrating the application of advanced machine learning models in a critical clinical setting, achieving significantly higher accuracy in blood loss estimation compared to traditional methods; (2) Validating the system's efficacy in real-time surgical environments, thereby enhancing the decision-making process and potentially reducing postoperative complications; (3) Setting a new standard in surgical care by integrating a complex system into real-world clinical workflows, showcasing its adaptability and potential for widespread adoption. Future work will focus on expanding the dataset and refining the algorithms to ensure the MDCare system's robustness and adaptability across surgical settings. The findings underscore the potential of MDCare to automate and refine critical aspects of surgery, marking a significant advancement in surgical care.