An approach to machine learning-based non-invasive hemoglobin estimation using multi-wavelength PPG signal features

The determination of hemoglobin (Hb) levels is pivotal in the clinical diagnosis and management of anemia and other blood disorders. Traditionally, Hb measurement requires blood samples, which is invasive and can cause patient discomfort and anxiety. There is a growing demand for non-invasive methods that can reduce patient stress and streamline the monitoring process. This study addresses this need by exploring the use of photoplethysmography (PPG) signals for non-invasive Hb measurement.

Utilizing raw PPG data from 68 subjects, which include both red and infrared light signals, this research applies feature extraction techniques. Three key features-mean, kurtosis, and skewness-were extracted from each signal type, as applied to the dataset, resulting in a comprehensive dataset of six features per subject. These features, along with demographic data such as gender and age, were used as inputs to a Multilayer Perceptron (MLP) neural network.

The neural network was adept at predicting Hb levels, achieving a Mean Relative Error (MRE) of less than 2.46%.

The implications of this research are significant, offering a potential shift in how blood hemoglobin levels are measured in clinical settings. By leveraging feature extraction methods and artificial neural networks, this study not only validates the efficacy of PPG as a non-invasive diagnostic tool but also paves the way for future advancements in medical technology. The successful application of machine learning techniques in this context highlights a pathway towards more patient-friendly, efficient, and accessible health monitoring systems.