Novel Analgesic Index for Postoperative Pain Assessment Based on a Photoplethysmographic Spectrogram and Convolutional Neural Network: Observational Study (Preprint)

Abstract

BACKGROUND Although commercially available analgesic indices based on biosignal processing have been used to quantify nociception during general anaesthesia, the performance of these indices is not high in awake patients. Therefore, there is a need for the development of a new analgesic index with improved performance to quantify postoperative pain in awake patients. OBJECTIVE The aim of this study was to develop a new analgesic index using spectrogram of photoplethysmogram and convolutional neural network to objectively assess pain in awake patients. METHODS Photoplethysmograms (PPGs) were obtained for 6 min both in the absence (preoperatively) and presence (postoperatively) of pain in a group of surgical patients. Of these, 5 min worth of PPG data, barring the first minute, were used for analysis. Based on the spectrogram from the photoplethysmography and convolutional neural network, we developed a spectrogram-CNN index (SCI) for pain assessment. The area under the curve (AUC) of the receiver-operating characteristic (ROC) curve was measured to evaluate the performance of the two indices. RESULTS PPGs from 100 patients were used to develop the SCI. When there was pain, the mean [95% confidence interval, CI] SCI value increased significantly (baseline: 28.5 [24.2 - 30.7] vs. recovery area: 65.7 [60.5 - 68.3]; P<0.01). The AUC of ROC curve and balanced accuracy were 0.76 and 71.4%, respectively. The cut-off value for detecting pain was 48 on the SCI, with a sensitivity of 68.3% and specificity of 73.8%. CONCLUSIONS Although there were limitations to the study design, we confirmed that the SCI can efficiently detect postoperative pain in conscious patients. Further studies are needed to assess feasibility and prevent overfitting in various populations, including patients under general anaesthesia. CLINICALTRIAL KCT0002080