Date of Award
Master of Science in Computer Science
Computer Science and Statistics
Krishna Kumar Venkatasubramanian
Physiological data can be used to detect the presence of pain, a problem that up to this point has entirely subjective solutions. While there are general indicators of pain, physiological signals have been shown to alter as a response to painful stimuli. Prior work has primarily focused on predicting a level of pain reported by a patient based on the assumption that pain is present. In this work, we present a means of using machine learning to identify the presence of pain using data collected from a freely available database MIMIC-III. Our methodology involves constructing an image reconstruction based classifier and evaluating our optimal classifiers on totally unseen testing data. Using both a 2 physiological stream and a 3 physiological stream approach, our models produced an accuracy of 80.56% and 87.18%, respectively. Each model is able to detect pain given less than a minute of data, although the 2 stream approach requires less data to work with. The proposed method for identifying pain presence has not been attempted before to our knowledge.
Jacobs, Derek, "PREDICTING PAIN PRESENCE IN ICU PATIENTS USING PHYSIOLOGICAL SIGNALS" (2022). Open Access Master's Theses. Paper 2152.