A Personalized, Uncertainty-Aware, Trustworthy Algorithm for Effective Pain Assessment using Biosignals

Abstract

Automatic pain assessment algorithms are used to improve pain assessment and assist subsequent pain treatment and management for patients without healthcare provider supervision. This thesis proposes a new pain assessment framework called "A Personalized, Uncertainty-Aware, Trustworthy Algorithm for Effective Pain Assessment using Biosignals." The framework takes into account the uncertainty of the data itself and the strong subjectivity of the pain experience, utilizing heart rate variability analysis to assess data uncertainty and test time adaptation to deal with distribution drift. It considers that pain data is imperfect, that there are data-label inconsistencies, and that the personalization of pain prediction algorithms is important. Our aim is to create complete frameworks for automated pain assessment that reduce the complexity of algorithms while predicting well. We collected experimental pain data and data from real pain patients, including post-surgical patients and women in labor. Through experiments and analyses, the framework outperforms state-of-the-art methods.