OxyMamba: Multi-Temporal Parallel Mamba for Camera-Based Blood Oxygen Estimation

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Contactless blood oxygen saturation (SpO$_{2}$) estimation is vital for health monitoring.
• Existing video-based methods struggle with long temporal dependencies, RGB spectral data, and physiological priors.

Purpose of the Study:

• To introduce OxyMamba, a state-space-model-based framework designed to overcome limitations in contactless SpO$_{2}$ estimation.
• To enhance the accuracy and reliability of SpO$_{2}$ measurement using video data.

Main Methods:

• Developed OxyMamba integrating a Multi-Temporal Parallel Mamba for temporal context.
• Incorporated a cross-channel complex-domain mixing module for RGB spectral analysis.
• Utilized an anchor-based heart-rate sideband gating mechanism for physiological plausibility.

Main Results:

• OxyMamba achieved low Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) on benchmark datasets (PURE, VIPL-HR, ViInHealth).
• The gating mechanism demonstrated performance improvements, particularly in low SpO$_{2}$ subgroups.
• Results indicate favorable performance within tested protocols.

Conclusions:

• OxyMamba presents a promising approach for contactless SpO$_{2}$ estimation.
• Further validation across diverse populations, conditions, and SpO$_{2}$ ranges is necessary before clinical deployment.