Background and Aims: Head movement has been known to be a significant impediment to using automated facial analysis to assess pain in young children. In this study we investigated the feasibility and validity of using automated single frame facial analysis to detect pain related facial action units (AUs) in infants 0-12 months undergoing a painful procedure (e.g. heel prick, immunisation).
Methods: The facial expressions of 30 infants were analysed 10 seconds after they underwent a painful procedure using PainChek® Infant app’s (iOS V1.1.15) automated recognition and facial analysis. Two different modes of analysis were used: standard 3 second video analysis (SVA), and adaptive video analysis (AVA) using 1, 2, 3 and 5 frames. All analyses were undertaken on video footage of the procedures, and all were repeated in triplicate. 
Results: Of 90 SVAs attempted 58 (64.4%) were completed successfully, compared with 346 of 360 (96.1%, p <0.0001) undertaken using the AVA. The success rates for the analyses using 1, 2, 3 and 5 frames were 96.7%, 96.7%, 95.6% and 95.6%, respectively. A significant difference in face score was seen only once in 72 (1.4%) occasions when SVA and AVA paired results could be compared. Similarly, only one of 108 (0.92%) comparisons of paired AVA results differed significantly.
Conclusions: These results demonstrate that the use of a single frame analysis provided comparable detection of pain related facial AU codes, when compared to SVA. Further, adaptive video analysis significantly increased the probability that a successful assessment could be completed.

PAIN ASSESSMENT IN INFANTS: INCREASING THE FEASIBILITY OF AUTOMATED FACIAL ASSESSMENT