AMS-HD: Hyperdimensional Computing for Real-Time and Energy-Efficient Acute Mountain Sickness Detection.

Area of Science:

• Biomedical Engineering
• Computer Science
• Altitude Medicine

Background:

• Acute mountain sickness (AMS) is a common illness at high altitudes, posing risks of severe health complications.
• Current machine learning methods for AMS detection lack the efficiency needed for continuous monitoring on wearable devices.

Purpose of the Study:

• To introduce AMS-HD, the first hyperdimensional computing (HDC) framework for real-time AMS detection.
• To develop a hardware-efficient solution for continuous AMS monitoring using wearable physiological signals.

Main Methods:

• Developed a complete HDC framework integrating feature selection, hypervector encoding, and positional projection.
• Implemented and validated the framework on ARM, FPGA, and smartwatch-smartphone platforms.
• Utilized wearable SpO2 and heart rate signals for AMS severity classification.

Main Results:

• AMS-HD achieved up to 91% accuracy and 90% F1-score in binary classification, outperforming SVM and MLP baselines.
• On FPGA, AMS-HD demonstrated significant reductions in LUT (7.3x) and flip-flop (5.8x) usage, with 3.9x lower power consumption compared to MLP.
• On mobile platforms, AMS-HD required minimal resources: 1% battery per session, 60 Bytes of memory, and 2.50 ms inference time.

Conclusions:

• AMS-HD offers a scalable, hardware-aware alternative for real-time AMS monitoring.
• The framework achieves competitive performance with substantially lower resource consumption than conventional ML methods.
• This work bridges wearable inference and low-level hardware deployment for resource-constrained health monitoring.