Few-shot learning improves personalized automated pain detection across subjects
Few-Shot Learning for Personalised Automated Pain Assessment
Machine LearningArtificial Intelligence
Summary
Pain is felt differently by each person, so it is hard to create one automated system that accurately measures pain for everyone. The authors tested a type of machine learning called few-shot learning, which quickly adapts to new people with only a few examples. This helped improve accuracy when identifying pain from signals in several datasets, even when testing on people not seen before. Their experiments show that personalizing pain classifiers using this technique can handle differences between individuals better.
What this means in practice
- •For medical device developers: Create pain monitoring devices that adapt to individual patients using few examples for improved assessment accuracy.$Commercial implications: Personalized pain assessment enables new adaptive medical devices tailored to patient variability, creating better pain management tools.
- •For clinical data teams: Implement personalized automated pain analysis systems that maintain accuracy when deployed on new patients without extensive retraining.
Authors
Heinke Hihn, Ibrahim Eisawy, Patrick Thiam, Hans A. Kestler, Friedhelm Schwenker
Abstract
Pain perception varies substantially across individuals, making it difficult for population-based classifiers to generalise across all subjects in a dataset. One way to account for subject variability is to train personalised classifiers. In this work, we evaluate Few-Shot Learning, a sub-area of Meta-Learning, as an approach to personalisation in automated pain assessment. We re-interpret the shift from population-level to subject-level evaluation as a task-domain shift, where the observed classes remain fixed but the target subject changes. We evaluate our method on the BioVid Pain Database, the SenseEmotion Database, and the PainMonit Experimental Dataset (PMED), reaching 85.75% and 35.49% accuracy on BioVid and 82.37% and 41.88% on SenseEmotion in the binary and multi-class settings under a Leave-One-Subject-Out CV protocol respectively, and 90.47% on PMED, for which only a binary benchmark exists. Using samples to implement k-shot conditioning, the accuracies can be improved to 86.25%, 40.06%, 83.43%, 44.08%, and 91.25%, respectively. To further evaluate the effects and robustness of our method, we provide additional ablation experiments and investigate the personalisation effects. Our results suggest that support-conditioned few-shot adaptation can improve average performance under inter-subject variability.