Automated Newborn Pain Assessment Framework Using Computer Vision Techniques

Pain evaluation in newborns is becoming a mandatory task in clinical practice. Currently, despite its complexity, pain assessment is entirely delegated to subjective estimates. The aim of this study is to propose an automated -- thus objective -- approach for neonatal pain evaluation focusing on facial expressions. From patients' face detection, a set of relevant parameters were extracted; both pixel-based image processing and analysis of facial landmarks led to final pain scores which were computed with respect to 3 widely adopted pain scales. The algorithm has been validated in a trial based on 15 videos acquired at the Neonatal Unit of AO Ordine Mauriziano Hospital, Turin, during heel stick procedure for blood sampling. The proposed algorithm scores have been compared to those subjectively assigned by health care professionals. The results confirm that manual pain assessment is a challenging task that often results in an elevated variance across scores between different operators, making automated evaluation highly desirable. The proposed algorithm is a first step in this direction, and despite difficulties in handling rapid facial changes, it is preparatory to the definition of an experimental protocol which merges video analysis, including the most relevant facial metrics, with audio processing for improved reliability.