LAST-PAIN: Learning Adaptive Spike Thresholds for Low Back Pain Biosignals Classification.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Spiking neural networks (SNNs) offer ultra-low-power computation potential on neuromorphic hardware.
• Efficient conversion of continuous data to discrete spike trains is a key challenge for SNNs.
• Existing encoding methods struggle with complex, high-dimensional data.

Purpose of the Study:

• Introduce Learning Adaptive Spike Thresholds (LAST), a trainable encoding strategy for SNNs.
• Enable efficient transformation of continuous, multi-dimensional data into sparse spike trains.
• Validate LAST's effectiveness in a healthcare application for biosignal analysis.

Main Methods:

• Developed the LAST encoder with adaptive thresholds for continuous signal transformation.
• Applied LAST to the EmoPain dataset for chronic lower back pain (CLBP) assessment.
• Compared LAST performance against conventional rate and latency-based encodings.

Main Results:

• LAST-driven SNN achieved 80.43% accuracy and 0.44 MCC for CLBP classification.
• The framework attained an F1-score of 0.65 in detecting protective behavior.
• LAST outperformed traditional encoding methods while maintaining sparse spike representations.

Conclusions:

• LAST provides an effective solution for encoding continuous biosignals into SNN-compatible spike trains.
• The LAST encoder shows promise for energy-efficient, real-time biosignal processing in resource-limited settings.
• This approach facilitates advanced SNN applications in healthcare and beyond.