A Multi-label Architecture for Vision-based Measurement of Intervals of Pain

Vision-based measurement of pain can provide significant advantage in patient care and cost reduction. The subjective nature of pain, however, poses metrological challenges yet to be addressed. With this work, we designed and validated a measurement architecture for automatically estimating intervals of pain over time based on the analysis of facial expressions. A reference measurement procedure was set up for labelling subjective levels of pain as observed by a set of independent evaluators. By means of a multi-label strategy, the system was calibrated for managing the uncertainty of the information provided by the evaluators at the ground-truth level. Results obtained for different coverage probability values support the efficacy of the proposed platform and motivate further investigations