STAL: Spike Threshold Adaptive Learning Encoder for Classification of Pain-Related Biosignal Data

TL;DR

This work introduces Spike Threshold Adaptive Learning (STAL), a trainable encoder that converts continuous sEMG and IMU biosignals into spike trains, enabling spiking neural networks to classify chronic lower back pain using the EmoPain dataset. An ensemble of Spiking Recurrent Neural Networks (SRNNs) processes multi-stream modalities (sEMG, Angle, Energy) with a Random Forest meta-classifier for final decision, addressing small sample size and class imbalance through targeted oversampling. The STAL-SRNN pipeline achieves strong performance (e.g., 80.43% accuracy, MCC 0.437, AUC 0.679), outperforming rate- and latency-based encoders and surpassing the best deep learning model in MCC, highlighting the potential of neuromorphic approaches for energy-efficient, wearable pain monitoring. The study advances neuromorphic biosignal analysis and lays groundwork for robust, real-time pain assessment and personalized rehabilitation using low-power hardware.

Abstract

This paper presents the first application of spiking neural networks (SNNs) for the classification of chronic lower back pain (CLBP) using the EmoPain dataset. Our work has two main contributions. We introduce Spike Threshold Adaptive Learning (STAL), a trainable encoder that effectively converts continuous biosignals into spike trains. Additionally, we propose an ensemble of Spiking Recurrent Neural Network (SRNN) classifiers for the multi-stream processing of sEMG and IMU data. To tackle the challenges of small sample size and class imbalance, we implement minority over-sampling with weighted sample replacement during batch creation. Our method achieves outstanding performance with an accuracy of 80.43%, AUC of 67.90%, F1 score of 52.60%, and Matthews Correlation Coefficient (MCC) of 0.437, surpassing traditional rate-based and latency-based encoding methods. The STAL encoder shows superior performance in preserving temporal dynamics and adapting to signal characteristics. Importantly, our approach (STAL-SRNN) outperforms the best deep learning method in terms of MCC, indicating better balanced class prediction. This research contributes to the development of neuromorphic computing for biosignal analysis. It holds promise for energy-efficient, wearable solutions in chronic pain management.

Figures (6)

• Figure 1: A schematic of the proposed architecture, including the Spike Threshold Adaptive Learning (STAL) encoder and Spiking Recurrent Neural Network (SRNN), which are responsible for classifying data from EmoPain aung2016EmoPain into Healthy and Chronic Lower Back Pain (CLBP) classes. The multimodal biosignals consist of sEMG sensors and derived features (i.e., Joint Energy and Joint Angle) from IMU sensors. These continuous signals convert into spike trains using the STAL encoder and are subsequently classified using an ensemble of SRNNs. The ensemble prediction is generated using a Random Forest meta classifier.
• Figure 2: The STAL architecture consists of feature extraction and feature-to-spike conversion modules. Here, $\mathbf{\mathcal{L}}$ is the encoder loss function, formulated in Eq. \ref{['eq:04']}.
• Figure 3: The SRNN architecture. The spike trains $\mathbf{\hat{B}}$ of each data type (sEMG, Energy, Angle) are classified separately. $T$ represent the total time steps, $c$ are the channels, $\tau$ is the time dimension of the Recurrent Leaky Integrate-and-Fire (R-LIF) neurons.
• Figure 4: (a) The influence of batch size on the F1-score of the proposed STAL-SRNN for each modality. (b) The Spike density for the proposed architecture as a function of $(\psi, p)$. While the difference in spike density between $(\psi, p) =(5,0.5)$ and $(\psi, p) =(10,0.5)$ is less than 1% (0.549 vs 0.540), the former achieves better performance across other classification metrics.
• Figure 5: Confusion matrices of the encoders with the SRNN classifier.