Accounting for Affect in Pain Level Recognition
This work addresses pain level recognition in real-world settings for healthcare applications, but it is incremental as it builds on existing datasets and methods.
The study tackled the problem of automated pain assessment by highlighting the importance of accounting for affect, showing that ignoring it degrades performance while incorporating it boosts recognition, with specific performance changes observed in simulations.
In this work, we address the importance of affect in automated pain assessment and the implications in real-world settings. To achieve this, we curate a new physiological dataset by merging the publicly available bioVid pain and emotion datasets. We then investigate pain level recognition on this dataset simulating participants' naturalistic affective behaviors. Our findings demonstrate that acknowledging affect in pain assessment is essential. We observe degradation in recognition performance when simulating the existence of affect to validate pain assessment models that do not account for it. Conversely, we observe a performance boost in recognition when we account for affect.