A novel combined deep learning methodology to non-invasively estimate hemoglobin levels in blood with high accuracy

Hemoglobin is an essential protein found in blood and should not fall below a certain level in humans. Today's methods of hemoglobin measurement are mostly invasive. This study aims to perform a non-invasive estimation of hemoglobin levels using age, height, weight, body mass index, gender, and nail images of individuals. Data was collected from 353 volunteers aged 1 to 92 years. Two different data sets were created using these data: a numerical dataset and a nail image set. A combined deep learning model was put forward using both the model created for numerical data and the model created for nail images. In this study, bias was calculated as 0.03 g/dL, and the limits of agreement value in the 95% confidence interval was calculated as 1.09 g/dL. The calculated mean absolute percentage error values were 2.09%, and the root mean squared error was 0.56 g/dL. After entering the necessary data into the system, the estimated average resulting time was 0.09 s. The results of this study have shown success compared to the results of similar studies, and this method can be used for non-invasive hemoglobin level estimation. The recommended approach is more comfortable than invasive methods and gives much faster results.