Physiological and Behavioral Profiling for Nociceptive Pain Estimation Using Personalized Multitask Learning
Date:
I presented a poster at the NeurIPS Machine Learning for Health (ML4H) Workshop 2017, describing our work on personalized pain estimation from multimodal data. The project introduced a method for building physiological and behavioral profiles based on individual responses to heat pain, and for using these profiles within a personalized multi-task neural network architecture.
As detailed in our paper, we extracted features from skin conductance, electrocardiography and facial behavior (including facial landmarks, head pose, eye gaze and facial action units) using data from the BioVid Heat Pain Database. We constructed subject specific descriptors summarizing physiological reactivity and facial expressiveness across pain levels, and applied normalized spectral clustering to group participants into pain response profiles. These profiles were then used to define the task structure of a multi task neural network for continuous pain estimation, enabling shared representations across similar subjects while adapting to inter individual variability.
The poster presented results showing that the personalized multi task model outperformed single modality and single task baselines across mean absolute error, root mean square error and intra class correlation metrics. The findings demonstrated that incorporating subject level profiles improves multimodal pain estimation performance and highlights the importance of personalization when modeling human affect and nociceptive responses.