A Non-Invasive Hemoglobin Detection Device Based on Multispectral Photoplethysmography.

Area of Science:

• Biomedical Engineering
• Clinical Diagnostics
• Optical Sensing Technologies

Background:

• Invasive hemoglobin measurement methods pose risks like infection and discomfort.
• Accurate hemoglobin monitoring is crucial for diagnosing and managing various clinical conditions.
• There is a need for convenient, non-invasive alternatives to traditional blood tests.

Purpose of the Study:

• To develop and validate a non-invasive hemoglobin detection method using multi-wavelength photoplethysmography (PPG).
• To design and implement a low-cost, user-friendly device for PPG signal acquisition.
• To evaluate the performance of different machine learning models for non-invasive hemoglobin estimation.

Main Methods:

• Utilized a custom-designed finger clip with eight LEDs (seven regular, one broadband) and sensors for transmissive PPG signal collection.
• Employed 3D printing for sensor integration, enabling simultaneous monitoring of LED-sensor distance and finger pressure.
• Extracted PPG signal features using sliding-window variance and applied AdaCost for signal evaluation, with AdaBoost, BPNN, and Random Forest models for regression.

Main Results:

• Pearson correlation analysis refined the dataset, highlighting the benefits of broadband LED light sources.
• The AdaBoost model demonstrated superior performance in non-invasive hemoglobin estimation.
• Achieved a mean absolute error (MAE) of 2.67 g/L and a correlation coefficient (R²) of 0.91 with the AdaBoost model.

Conclusions:

• The developed device effectively enables non-invasive hemoglobin detection.
• This novel methodological approach offers a promising alternative for clinical hemoglobin measurements.
• The findings support the potential application of this technology in clinical settings for improved patient care.