Machine learning for automated pain assessment using physiological signals

Abstract

Pain assessment is a critical step in diagnosing, choosing an intervention, and patient monitoring. In current clinical settings, typically pain is measured by a patient's self-reported information using the Visual Analog Scale (VAS) and Numerical Rating Scale (NRS). This subjective pain assessment results in suboptimal treatment plans, over-prescription of opioids, and drug-seeking behavior among patients. Moreover, these methods have severe limitations when patients cannot validly communicate their pain intensity, especially infants, toddlers, or adults with mental disabilities. Hence, objective measurement of pain has long been clinicians' goal for effective pain management. The first study explored automatic objective pain intensity estimation machine learning models using inputs from physiological sensors. We extracted 66 features from Electrodermal Activity (EDA), Electrocardiogram (ECG), Electromyogram (EMG) signals collected from study participants subjected to heat pain. We built different machine learning models, including Support Vector Regression, Neural Networks, and Extreme Gradient Boosting. Then we identified the best combination of physiological sensor, feature set, and machine learning model that gives the best predictive performance. Then, a clustering-based-SVR model was proposed to address the inter-subject variabilities in physiological responses to pain. The second study applied machine learning techniques on Blood Volume Pulse (BVP) signals to device a non-invasive modality for pain sensing. Thirty-two healthy subjects participated in the Cold Pressor Test. We investigated a novel set of time-domain, frequency-domain, and nonlinear dynamics features that could potentially be sensitive to pain. We identified 24 features from BVP signals and 20 additional features from Inter-beat Intervals (IBIs) derived from the same BVP signals to be useful for pain prediction. Our results suggest that BVP signal together with machine learning algorithms is a promising physiological measurement for automated pain assessment. The third study explored the Bidirectional LSTM Recurrent Neural Networks on EDA signals from the cold pressor test. This study proposed an ensemble of Bi-LSTM RNN and Extreme Gradient Boosting Decision Trees (XGB) which gave very good model performance for multiclass pain intensity classification. It also explored a hybrid concatenation of the deep-learning representations and a set of 14 handcrafted features. This study showed that deep learning representations go beyond expert knowledge in extracting features from physiological signals for objective and accurate pain assessment. --Author's abstract