Skin Conductance Deconvolution for Pain Estimation
Date:
I presented a poster at the International Conference on Biomedical and Health Informatics (BHI 2018) in Las Vegas, describing our work on estimating pain intensity from skin conductance signals. The project, conducted at the MIT Media Lab, focused on leveraging noninvasive physiological sensing to quantify nociceptive responses when self-report is not feasible.
The work introduced a method for skin conductance deconvolution based on nonnegative deconvolution to recover the underlying sudomotor nerve activity. Using the BioVid Heat Pain database, we extracted features from the phasic driver signal (including burst count, reconvolved amplitude, tonic level, average and maximum phasic activity, and integrated phasic response) to distinguish pain trials from baseline. We evaluated binary classifiers across multiple pain levels and observed stronger autonomic signatures at higher pain intensities, while lower pain levels remained more difficult to detect.
