Enhancing non-invasive blood loss measurement through real-time video processing

Abstract

Blood loss during surgery remains a critical concern, as maintaining appropriate Packed Cell Volume (PCV) levels is essential for patient safety. However, challenges in blood preparation and compatibility, due to varying blood types, often complicate timely and effective management. The aim of this study is to demonstrate the potential of enhancing non-invasive blood loss measurement through real-time video processing with machine learning integration. This was achieved by developing a non-invasive system that estimates hemoglobin levels using real-time video recordings of participants' index fingers. The system employs image processing techniques and machine learning algorithms to analyze color variations correlated with blood characteristics. This approach effectively bridges a significant gap in traditional blood loss measurement methods, which often lack accuracy, objectivity, or real-time capabilities. Using video processing techniques and feature extraction methods, such as the Erythema Index (EI) and Red Intensity (RI), alongside robust regression models, the system was able to achieved high predictive accuracy (MAPE: 1.89%, R²: 93.35%). The findings demonstrate that reliable hemoglobin estimation can be achieved using simple video inputs, making this solution especially valuable for resource-limited healthcare settings. It paves the way for broader clinical adoption of low cost, real-time monitoring tools that enhance patient safety during surgical and emergency procedures.