Аннотация:Accurate and rapid assessment of hemoglobin (Hb) concentration is essential for medical decision-making in preoperative evaluations, anemia screening, and blood loss monitoring. RGB imaging, which estimates Hb levels by analyzing tissue color in regions such as nail beds, has attracted growing interest due to its portability and compatibility with mobile devices. While promising, the reported prediction errors vary widely from 5.6g/L to 36g/L, raising concerns about clinical applicability. External physiological perturbations can significantly affect the reliability of RGB-based Hb estimation by altering tissue optical properties and introducing measurement bias. However, their systematic influence on signal quality and prediction accuracy remains insufficiently explored. In this study, we systematically evaluate the effects of ambient temperature, mechanical compression, and venous occlusion on RGB-based Hb estimation using a machine learning model trained on data from 298 patients with matched venous Hb values. Controlled experiments reveal that these external conditions introduce systematic biases ranging from 5g/L to 20g/L. To address intersubject variability, we propose a personalized intercalibration approach based on prior Hb measurements from the same individual, which we validate on 17 repeat blood donors. These findings provide a comprehensive quantification of how external and individual-specific factors affect non-invasive Hb prediction, and support the development of accurate, robust, and clinically viable RGB-based monitoring systems for both point-of-care and mobile healthcare applications.
