Machine Learning for Pain Medicine: Physiological and Behavioral Profiling for Nociceptive Pain Estimation

I presented a poster at the Harvard–MIT Health Sciences and Technology (HST) Forum at Harvard Medical School, describing my research on personalized machine learning approaches for estimating nociceptive pain. The work, conducted at the MIT Media Lab, explored how individual differences in physiological and behavioral responses to pain can be leveraged to improve continuous pain intensity estimation.

The poster introduced a method for building subject specific pain response profiles by applying spectral clustering to multimodal features, including skin conductance, electrocardiography and facial expressiveness. As shown in the poster’s profiling section (upper right of the poster), clusters were constructed from average physiological and behavioral descriptors computed during the first 48 heat pain stimuli, enabling groups of subjects with similar response patterns to be identified.

Using these profiles, we trained a clustered multi task neural network for continuous pain estimation from overlapping six second windows of multimodal data. The model architecture shared early network layers across all subjects and used profile specific layers to capture heterogeneity across individuals. The approach was evaluated on the BioVid Heat Pain Database and produced improved performance relative to single task and single modality baselines.

This poster summarized how combining multimodal sensing, subject profiling and multi task learning can enhance automatic pain estimation, highlighting the importance of personalization in modeling human affect and nociceptive responses.