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

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Machine Learning in Healthcare

Background:

• Traditional hemoglobin (Hb) measurement involves invasive blood sampling, causing patient discomfort.
• Non-invasive methods for Hb monitoring are in demand to improve patient experience and streamline clinical workflows.
• Photoplethysmography (PPG) offers a potential non-invasive approach using light signals.

Purpose of the Study:

• To explore the efficacy of photoplethysmography (PPG) signals for non-invasive hemoglobin (Hb) measurement.
• To develop and validate a machine learning model for predicting Hb levels from PPG data.
• To provide a patient-friendly alternative to traditional blood-based Hb testing.

Main Methods:

• Collected raw PPG data (red and infrared light signals) from 68 subjects.
• Extracted statistical features (mean, kurtosis, skewness) from PPG signals.
• Utilized a Multilayer Perceptron (MLP) neural network with extracted features and demographic data for Hb prediction.

Main Results:

• The MLP neural network achieved a Mean Relative Error (MRE) of less than 2.46% in predicting Hb levels.
• The study demonstrates the feasibility of using PPG-derived features for accurate Hb estimation.
• Feature extraction and machine learning effectively processed PPG data for clinical application.

Conclusions:

• PPG signals, combined with feature extraction and artificial neural networks, provide a viable non-invasive method for Hb measurement.
• This approach has significant implications for clinical diagnostics, offering a more comfortable and efficient monitoring solution.
• The study highlights the potential of machine learning in advancing accessible and patient-centric healthcare technologies.