Deep BCI of Pain Decoding from fNIRS

Abstract

BCI is a hybrid human artificial intelligence system that promotes physical or cognitive augmentation by artificial intelligence decoding human neurological behavior and feeding it back to humans through artifacts such as robots or computers. This study proposes Deep BCI to alleviate the patient’s pain by objectively determining the intensity of pain. This paper deals with the neural decoding of pain, part of Deep BCI. We present a deep learning method to specify pain conditions from the neurological features induced by thermal pain stimulation. We established a thermal stimulation experimental set-up by international standard thermal QST and adopted fNIRS to measure neurological features. An LSTM model was trained to accurately extract fNIRS features associated with the perceived nociceptive pain intensity. As a proof of concept, we applied this trained LSTM model to classify the boundary between pain and non-pain. The accuracy of the classifier was 96.95% for the cold pain vs. non-pain and 96.90% for the hot pain vs. non-pain. Based on this proof-of-concept result, we will develop artificial intelligence that predicts pain levels and applies it to Deep BCI for pain relief treatment.