Machine Learning-based Deep Analysis of Human Blood using NIR Spectrophotometry Signatures

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Machine Learning

Background:

• Non-invasive diagnostics are crucial for patient safety.
• Anemia detection using spectroscopic and image analysis shows promise but faces challenges with inconsistent results.
• Developing reliable, non-invasive hemoglobin monitoring is essential.

Purpose of the Study:

• To present an adaptable point-of-care Near-Infrared (NIR) spectrophotometric approach for hemoglobin (Hb) concentration monitoring.
• To integrate a machine learning (ML) algorithm that accounts for influencing factors.
• To improve the accuracy and consistency of non-invasive anemia detection.

Main Methods:

• An observational study involving 121 subjects with a wide range of Hb concentrations (8.2-17.4 g/dL).
• Utilized Near-Infrared (NIR) spectrophotometry and data from two standard laboratory analyzers.
• Applied dimensionality reduction techniques and regression models with 5-fold cross-validation.

Main Results:

• Support Vector Regression (SVR) with mutual information achieved optimal accuracy (r=0.79, SD=1.07 g/dL, bias=-0.13 g/dL).
• High comparability was found between the NIR system and laboratory analyzers (r=0.97, SD=0.50 g/dL, bias=0.21 g/dL).
• The system demonstrated precision of ±1 g/dL, irrespective of patient demographics and lifestyle factors.

Conclusions:

• The NIR spectrophotometric system with an ML algorithm provides precise and reliable hemoglobin monitoring.
• This technology is suitable for continuous Hb monitoring in remote rural areas and critical care units.
• The developed approach addresses the need for consistent and non-invasive anemia detection.