Hybrid RNN-ANN Based Deep Physiological Network for PainRecognition

Abstract

Quantitative assessment of pain is vital progressin treatment choosing and distress relief for patients. However,previous approaches based on self-report fail to provide objec-tive and accurate assessments. For impartial pain classificationbased on physiological signals, a number of methods have beenintroduced using elaborately designed handcrafted features. Inthis study, we enriched the methods of physiological-signal-based pain classification by introducing deep Recurrent NeuralNetwork (RNN) based hybrid classifiers which combines auto-extracted features with human-experience enabled handcraftedfeatures. A bidirectional Long Short-Term Memory network(biLSTM) was applied on time series of pre-processed signalsto automatically learn temporal dynamic characteristics fromthem. The handcrafted features were extracted to fuse withRNN-generated features. Finely selected features from biLSTMlayer output and handcrafted features trained an ArtificialNeural Network (ANN) to classify the pain intensity. The hand-crafted features enhance the RNN classification performanceby complementing RNN-generated features. With our accuracyreaching 83.3%, comparison results on an open dataset withother methods show that the proposed algorithm outperformsall of the previous researches with higher classification accuracy.Therefore, this research is a good demonstration of introducinghybrid features for pain assessment.

Run Wang

Master in Information Technology and Electrical Engineering

Formerly with extensive experience in algorithm design, now transitioned into the realm of digital architecture IC design.