Assessing pain objectively is challenging, as people vary considerably in how they experience
and express pain. Automated facial expression analysis offers an approach for assessing painrelated facial behavior from video. Expressed pain can be accurately detected using the AFFDEX Pain metric. The model achieves a ROC-AUC of 96.1% on an internal dataset comprising both a short-lasting pain paradigm based on the cold pressor test and a longer-lasting pain paradigm based on delayed onset muscle soreness, exceeding several recently reported ROC-AUC results on the popular UNBC dataset and other pain datasets evaluated in recent scientific publications.
However, facial expressions show intersubject variability. Thus, adding physiological measures
such as GSR or ECG allows for cross-validation, mitigating potential mis-identifications that may arise from facial masking and low facial expressivity. Combining physiological measurements with facial expression analysis can give unique and novel insights into the subjective perception and variable expression of pain.
Download to discover how combining automated facial expression analysis with physiological sensor data can provide reliable, multimodal insights into the subjective perception and variable expression of pain.
