Title:Pain Assessment based on fNIRS using Bidirectional LSTMs

Abstract:Assessing pain in patients unable to speak (also called non-verbal patients) is extremely complicated and often is done by clinical judgement. However, this method is not reliable since patients vital signs can fluctuate significantly due to other underlying medical conditions. No objective diagnosis test exists to date that can assist medical practitioners in the diagnosis of pain. In this study we propose the use of functional near-infrared spectroscopy (fNIRS) and deep learning for the assessment of human pain. The aim of this study is to explore the use deep learning to automatically learn features from fNIRS raw data to reduce the level of subjectivity and domain knowledge required in the design of hand-crafted features. Four deep learning models were evaluated, multilayer perceptron (MLP), forward and backward long short-term memory net-works (LSTM), and bidirectional LSTM. The results showed that the Bi-LSTM model achieved the highest accuracy (90.6%)and faster than the other three models. These results advance knowledge in pain assessment using neuroimaging as a method of diagnosis and represent a step closer to developing a physiologically based diagnosis of human pain that will benefit vulnerable populations who cannot self-report pain.