Automated Nociceptive Pain Assessment Using Physiological Signals and a Hybrid Deep Learning Network

Postoperative pain management has become a major medical and nursing challenge. Nowadays, hospitals have taken initiatives to measure acute pain using self-report measures like the Visual Analogue Scale and Numeric Pain Intensity Scale. But these methods are inaccurate as it depends on patient's input. Therefore, there is a requirement for an objective, quantitative method to monitor pain continuously. In this work, an automated acute nociceptive pain recognition system was proposed to objectively measure nociceptive pain using physiological signals and a hybrid Deep Learning network. The hybrid deep learning network constitutes a shallow CNN network that extracts the essential information of the pain from the physiological signals and the extracted feature matrix is fed to the LSTM network for feature concatenation. This process realizes the mapping of nociceptive pain from input data to detection. This work utilizes the BioVid Heat Pain database. A Unimodal hybrid CNN_LSTM network (using ECG signals) has achieved 68.70 percent, 62.61 percent, 67.86 percent, and 75.21 percent of classification accuracy for classification events (BL1 Vs PA1, BL1 Vs PA2, BL1 Vs PA3, and BL1 Vs PA4). Similarly, for classification events (BL1 Vs PA1, BL1 Vs PA2, BL1 Vs PA3, and BL1 Vs PA4), the unimodal hybrid CNN_LSTM network (using EDA signals) achieved 85.65 percent, 74.47 percent, 80.80 percent, and 80.17 percent of classification accuracy. Finally, for classification events (BL1 Vs PA1, BL1 Vs PA2, BL1 Vs PA3, and BL1 Vs PA4), the CNN_LSTM multimodal hybrid network (using both ECG and EDA signals) achieved 93.91 percent, 86.97 percent, 90.75 percent, and 94.12 percent of the classification accuracy, respectively.