Decoding Blood Volume Loss: Can Non-Invasive Signals Match the Gold Standard?

Area of Science:

• Physiological signal processing
• Medical diagnostics
• Machine learning in healthcare

Background:

• Hypovolemia diagnosis relies on invasive methods or clinical signs.
• Advanced waveform analysis of physiological signals for hypovolemia is underexplored.
• Deep learning (DL) offers potential for analyzing complex physiological data.

Purpose of the Study:

• To compare the diagnostic ability of invasive (ABP) and non-invasive (PPG) signals for classifying hypovolemia levels.
• To apply a deep learning framework for analyzing waveform data.
• To evaluate the performance of DL models in detecting mild, moderate, and severe hypovolemia.

Main Methods:

• Simulated hypovolemia using lower body negative pressure (LBNP) in healthy volunteers.
• Acquired and analyzed invasive arterial blood pressure (ABP) and non-invasive photoplethysmography (PPG) waveforms.
• Developed a supervised deep learning model using time-frequency representations and late fusion for ternary classification of hypovolemia levels.

Main Results:

• Both PPG and ABP signals could detect varying levels of hypovolemia.
• The non-invasive PPG signal achieved comparable or superior classification performance to ABP.
• PPG demonstrated high diagnostic potential with an AUROC of 0.8861 and F1 score of 0.7216.

Conclusions:

• Advanced waveform analysis using deep learning is effective for diagnosing hypovolemia.
• Photoplethysmography (PPG) offers a promising non-invasive alternative for hypovolemia detection.
• This approach has significant potential for real-time clinical monitoring and early diagnosis.