Practical Considerations of a BMI Application for Detecting Acute Pain Signals

Sile Hu, Zhengdong Xiao, Qiaosheng Zhang, Louise Urien, Jing Wang, Zhe Chen

Abstract: Brain-machine interfaces (BMIs) have been an important research area in closed-loop neuroscience and neuroengineering. In real-time neuroscience applications, many issues require special consideration, such as trial variability, spike sorting noise or multi-unit activity. For a BMI application of detecting acute pain signals, we discuss several practical issues in BMI applications and propose a new approach for change-point detection based on ensembles of independent detectors. Motivated from unsupervised ensemble learning, the “ensembles of change-point detectors” (ECPDs) combine the decision results from multiple independent detectors, which may be trained from data recorded at different trials or derived from different methodologies. The goal of ECPDs is to reduce the detection error (in terms of false negative and false positive rates) in online BMI applications. We validate our method using computer simulations and experimental recordings from freely behaving rats.