Non-invasive Haemoglobin Estimation by Observing Nail Color: A PCA based Approach

Abstract

Although conventional laboratory based haemoglobin estimation provides accurate estimation, however, it is time-consuming, painful, inconvenient in cases of chronic anaemic patients for regular diagnoses. Moreover, it needs trained professionals with specialized laboratory equipment. This paper presents a haemoglobin prediction model non-invasively by analysing the videophotography of color changes in finger nail-bed caused due to certain amount of pressure application and sudden release. To estimate the haemoglobin level, three most relevant frames are considered that comprise significant changes in intensity values due to blood occlusion and release. A reduced dimensional feature set is introduced by applying Principal Component Analysis (PCA) to enhance the coordination between the feature values. The reduced dimensional features are correlated with the clinically validated haemoglobin level through an ensemble prediction module comprising of a set of regression model and Multilayer Perceptron (MLP) architecture for predicting the haemoglobin level of unlabeled samples. The proposed system proves its efficiency with respect to the Root Mean Square Error (RMSE) value, prediction accuracy, classification accuracy, specificity and sensitivity.